< Dongguk Open Databases & CNN Model >

 

---------------------------------------------

33. Dongguk DenseNet-based Finger-vein Recognition Model (DDFRM) with algorithms

32. Dongguk OR-Skip-Net Model for Image Segmentation with Algorithm and Black Skin People (BSP) Label Information

31. Dongguk Banknote Type and Fitness Database (DF-DB3) & CNN Model with algorithms

30. Dongguk RetinaNet for Detecting Road Marking Objects with Algorithms and Annotated Files for Open Databases

29. Dongguk CNN Model for NIR Ocular Recognition (DC4NO) with algorithm

28. Dongguk Face Presentation Attack Detection Algorithms by Spatial and Temporal Information (DFPAD-STI)

27. Dongguk Dual Camera-based Driver Database (DDCD-DB1) and Trained Faster R-CNN Model with Algorithm

26. Dongguk FRED-Net with Algorithm

25. Dongguk Face and Body Database (DFB-DB1) with CNN models and algorithms

24. Dongguk Night-Time Face Detection database (DNFD-DB1) and algorithm including CNN model

23. Dongguk Iris Spoof Detection CNN Model version 2 (DFSD-CNN-2) with Algorithm

22. Dongguk Fitness Database (DF-DB2) & CNN Model

21. Dongguk-body-movement-based human identification

database version 2 (DBMHI-DB2) & CNN Model

20. Dongguk Multimodal Recognition CNN of Finger-vein and Finger shape (DMR-CNN) with Algorithm

19. Dongguk Drone Camera Database (DDroneC-DB2) with CNN models

18. Dongguk Periocular Database (DP-DB1) with CNN models and algorithms

17. Dongguk IrisDenseNet CNN Model (DI-CNN) with Algorithm

16. Dongguk Visible Light Iris Recognition CNN Model (DVLIR-CNN)

15. Dongguk Aggressive and Smooth Driving Database (DASD-DB1)

and CNN Model

14. Dongguk Night-time Pedestrian Detection Faster R-CNN and

Algorithm

13. Dongguk Shadow Detection Database (DSDD-DB1) & CNN

Model

12. Dongguk driver gaze classification database (DDGC-DB1) and

CNN model

11. Dongguk Age Estimation CNN Model (DAE-CNN)

10. Dongguk Single Camera-based Driver Database (DSCD-DB1)

9. ISPR Database (real and presentation attack finger-vein images)

& Algorithm Including CNN Model

8. Dongguk Visible Light & FIR Pedestrian Detection Database

(DVLFPD-DB1) & CNN Model

7. Dongguk Open and Closed Eyes Database (DOCE-DB1) & CNN

Model

6. Dongguk Multi-national Currencies Database (DMC-DB1) & CNN

Model

5. Dongguk Finger-Vein Database (DFingerVein-DB1) & CNN Model

4. Dongguk Night-time Human Detection Database (DNHD-DB1) &

CNN Model

3. Dongguk Body-based Person Recognition Database (DBPerson-

Recog-DB1)

2. Dongguk Body-based Gender Database (DBGender-DB2)

1. Dongguk Activities & Actions Database (DA&A-DB1)

 

 

33. Dongguk DenseNet-based Finger-vein Recognition Model (DDFRM) with algorithms

 

(1) Introduction

We trained finger-vein recognition system based on DenseNet-161 using two databases including the Hong Kong Polytechnic University Finger Image Database (version 1) [1] and the Shandong University homologous multi-modal traits (SDUMLA-HMT) finger-vein database [2]. We made trained models/Algorithm open to other researchers.

 

1. Kumar, A.; Zhou, Y. Human identification using finger images. IEEE Trans. Image Process. 2012, 21, 2228-2244.

2. SDUMLA-HMT Finger Vein Database. Available online: http://mla.sdu.edu.cn/info/1006/1195.htm (accessed on 7 May 2018).

 

(2) DDFRM with Algorithm Request

To gain access to the models and algorithm, download the following request form for DDFRM with algorithm. Please sign and scan the request form and email to Mr. Jong Min Song (whdwhd93@gmail.com).

 

Any work that uses this CNN model must acknowledge the authors by including the following reference.

 

J. M. Song, W. Kim, and K. R. Park, Finger-vein Recognition Based on Deep DenseNet Using Composite Image and Shift Matching, IEEE Access, in submission.

< Request Form for DDFRM with algorithm >

 

Please complete the following form to request access to the DDFRM with algorithm (All contents must be completed). This CNN model with algorithm should not be used for commercial use.

 

Name:

 

Contact:  (Email)

(Telephone)

 

Organization Name:

 

Organization Address:

 

Purpose:

 

 

Date:

 

                Name (signature)

 

 

 

32. Dongguk OR-Skip-Net Model for Image Segmentation with Algorithm and Black Skin People (BSP) Label Information

 

(1) Introduction

We trained outer skip connection-based deep convolutional network (OR-Skip-net) for image segmentation related to medical diagnosis and other applications, to evaluate the segmentation performance system using ten databases including HGR [1], EDds [2], LIRIS [2], SSG [2], UT [2], AMI [2], Pratheepan [3], BSP, Warwick-QU [4], and NICE.II [9]. We made trained models/Algorithm and BSP label information open to other researchers.

 

1.     Hand detection and pose estimation for creating human-computer interaction project. Available online: http://sun.aei.polsl.pl/~mkawulok/gestures/ip.html (accessed on October 31, 2018).

2.     Skin detection datasets for video monitoring. Available online: http://www-vpu.eps.uam.es/publications/SkinDetDM/ (accessed on November 5, 2018).

3.     Pratheepan dataset + ground truth. Available online: http://cs-chan.com/downloads_skin_dataset.html (accessed on November 5, 2018).

4.     GlaS@MICCAI'2015: Gland segmentation challenge contest. Available online: https://warwick.ac.uk/fac/sci/dcs/research/tia/glascontest/ (accessed on 24 January 2019).

5.     NICE. II. Noisy iris challenge evaluation - part II. Available online: http://nice2.di.ubi.pt/ (accessed on November 8, 2018).

 

(2) Request for OR-Skip-Net Model with Algorithm and Black Skin People (BSP) Label Information

To gain access to the models, algorithm and BSP label information, download the following request form. Please sign, scan the request form, and email to Mr. Arsalan (arsal@dongguk.edu).

Any work that uses this CNN model with algorithm and label information must acknowledge the authors by including the following reference.

 

Muhammad Arsalan, Dong Seop Kim, Muhammad Owais, and Kang Ryoung Park, OR-Skip-Net: Outer Residual Skip Network for Skin Segmentation in Non-Ideal Situations, Expert Systems with Applications, in submission.

 

 

< Request form for OR-Skip-Net model with algorithm

and BSP label information >

 

Please complete the following form to request access to the OR-Skip-Net model with algorithm and BSP label information (All contents must be completed). This CNN model with algorithm and label information should not be used for commercial use.

 

Name:

 

Contact:  (Email)

(Telephone)

 

Organization Name:

 

Organization Address:

 

Purpose:

 

 

Date:

 

                Name (signature)

 

 

 

31. Dongguk Banknote Type and Fitness Database (DF-DB3) & CNN Model with algorithms

 

(1) Introduction

We make Dongguk Banknote Type and Fitness Database (DF-DB3) based on Indian rupee (INR) (INR10/20/50/100/500/1000), the Korean won (KRW) (KRW 1000/5000/10000/50000) and United States dollar (USD) (USD 5/10/20/50/100) banknotes, and trained CNN Models of AlexNet, GoogleNet, ResNet-18/50 with algorithms available for the fair comparison by other researchers.

 

(2) Request for Dongguk Banknote Type and Fitness Database (DF-DB3) & CNN Model with algorithms

To gain access to these files, download the following request form. Please scan the request form and email to Dr. Tuyen Danh Pham (phamdanhtuyen@dongguk.edu). Any work that uses these files with algorithm must acknowledge the authors by including the following reference.

 

Tuyen Danh Pham, Dat Tien Nguyen, Chanhum Park and Kang Ryoung Park, Deep Learning-based Multinational Banknote Type and Fitness Classification with the Combined Images by Visible-Light Reflection and Infrared-Light Transmission Image Sensors, Sensors, in submission.

 

===========================================================================================================================================================================================================

 

< Request Form for Dongguk Banknote Type and Fitness Database (DF-DB3) & CNN Model with algorithms >

 

Please complete the following form to request access to these files (All contents must be completed). These files should not be used for commercial use.

 

Name :

 

Contact : (Email)

(Telephone)

 

Organization Name :

 

Organization Address :

 

Purpose :

 

 

Date :

 

               Name (signature)

 

 

 

30. Dongguk RetinaNet for Detecting Road Marking Objects with Algorithms and Annotated Files for Open Databases

 

(1) Introduction

Although the open databases of the Malaga urban dataset [1], the Daimler dataset [2], and the Cambridge dataset [3] have been widely used in previous studies, they do not provide annotated information of road markings. This increases the time and load for system implementation. Therefore, we provide the manually annotated information of road markings for the Malaga urban dataset, the Daimler dataset, and the Cambridge dataset. We also provide the proposed RetinaNet models trained by these databases based on different backbones with and without pre-trained weights to other researchers.

 

1. The Málaga Stereo and Laser Urban Data Set MRPT. Available online: https://www.mrpt.org/MalagaUrbanDataset (accessed on 1 October 2018).

2. Daimler Urban Segmentation Dataset.

Available online: http://www.6d-vision.com/scene-labeling (accessed on 1 October 2018).

3. Cambridge-driving Labeled Video Database (CamVid). Available online: http://mi.eng.cam.ac.uk/research/projects/VideoRec/CamVid/ (accessed on 1 October 2018).

 

(2) Request for Dongguk RetinaNet for Detecting Road Marking Objects with Algorithms and Annotated Files for Open Databases

To gain access to these files, download the following request form. Please scan the request form and email to Mr. Toan Minh Hoang (hoangminhtoan@dongguk.edu). Any work that uses these files with algorithm must acknowledge the authors by including the following reference.

 

Toan Minh Hoang, Phong Ha Nguyen, Noi Quang Truong, Young Won Lee, and Kang Ryoung Park, Deep RetinaNet-based Detection and Classification of Road Markings by Visible Light Camera Sensor, Sensors, in submission.

 

===========================================================================================================================================================================================================

 

< Request Form for Dongguk RetinaNet for Detecting Road Marking Objects with Algorithms and Annotated Files for Open Databases >

 

Please complete the following form to request access to these files (All contents must be completed). These files should not be used for commercial use.

 

Name :

 

Contact : (Email)

(Telephone)

 

Organization Name :

 

Organization Address :

 

Purpose :

 

 

Date :

 

               Name (signature)

 

 

 

29. Dongguk CNN Model for NIR Ocular Recognition (DC4NO) with algorithm

 

(1) Introduction

We made our algorithm for rough pupil detection based on sub-block based template matching and deep ResNet models trained with three open databases: CASIA-Iris-Distance, CASIA-Iris-Lamp, and CASIA-Iris-Thousand [1]. We made these trained CNN models for ocular recognition with algorithm open to other researchers.

 

1. CASIA-iris version 4. Available online:

 http://www.cbsr.ia.ac.cn/china/Iris%20Databases%20CH.asp (accessed on 9 November 2018)

 

(2) Request for DC4NO with algorithm

To gain access to the DC4NO with algorithm, download the following request form. Please scan the request form and email to Mr. Young Won Lee (lyw941021@dongguk.edu).

Any work that uses this DC4NO with algorithm must acknowledge the authors by including the following reference.

 

Young Won Lee, Ki Wan Kim, Toan Minh Hoang, Muhammad Arsalan, and Kang Ryoung Park, Deep Residual CNN-based Ocular Recognition Based on Rough Pupil Detection in the Images by NIR Camera Sensor, Sensors, in submission.

 

===========================================================================================================================================================================================================

 

< Request Form for DC4NO with algorithm >

 

Please complete the following form to request access to the DC4NO with algorithm (All contents must be completed). This DC4NO with algorithm should not be used for commercial use.

 

Name :

 

Contact : (Email)

(Telephone)

 

Organization Name :

 

Organization Address :

 

Purpose :

 

 

Date :

 

               Name (signature)

 

 

 

28. Dongguk Face Presentation Attack Detection Algorithms by Spatial and Temporal Information (DFPAD-STI)

 

(1) Introduction

We made stacked convolutional neural network (CNN)-recurrent neural network (RNN) along with handcrafted features for face presentation attack detection using the images from CASIA database [1] and Replay-mobile dataset [2], respectively. We made these trained CNN model open to other researchers.

 

1. Zhang, Z.; Yan, J.; Liu, S.; Lei, Z.; Yi, D. Li, S. Z. A face anti-spoofing database with diverse attack. In Proceedings of the 5th International Conference on Biometric, New Delhi, India, 29 March 1 April, 2012.

2. Costa-Pazo, A.; Bhattacharjee, S.; Vazquez-Fernandez, E.; Marcel, S. The replay-mobile face presentation attack database. In Proceedings of the International Conference on the Biometrics Special Interest Group, Darmstadt, Germary, 21-23 September, 2016.

 

(2) Request for DFPAD-STI

To gain access to the DFPAD-STI, download the following request form. Please scan the request form and email to Prof. Dat Tien Nguyen (nguyentiendat@dongguk.edu).

Any work that uses this DFPAD-STI must acknowledge the authors by including the following reference.

 

Dat Tien Nguyen, Tuyen Danh Pham, Min Beom Lee, and Kang Ryoung Park, Visible-Light Camera Sensor-based Presentation Attack Detection for Face Recognition by Combining Spatial and Temporal Information, Sensors, in submission.

 

===========================================================================================================================================================================================================

 

< Request Form for DFPAD-STI >

 

Please complete the following form to request access to the DFPAD-STI (All contents must be completed). This DFPAD-STI should not be used for commercial use.

 

Name :

 

Contact : (Email)

(Telephone)

 

Organization Name :

 

Organization Address :

 

Purpose :

 

 

Date :

 

               Name (signature)

 

 

 

27. Dongguk Dual Camera-based Driver Database (DDCD-DB1) and Trained Faster R-CNN Model with Algorithm

 

(1) Introduction

When acquiring DDCD-DB1, the drivers gaze area was divided into 15 zones. The drivers gazed at the 15 zones divided out beforehand in order, and a total of 26 participants were each assigned 8 different situations (i.e., wearing a hat, wearing four different types of glasses (rimless, gold-rimmed, half-frame, and horn-rimmed), wearing sunglasses, making a call through mobile phone, covering face through hand, etc.), and the data were collected by two NIR cameras with NIR illuminators. As the participants gazed at the designated regions in turn, natural head rotations that would occur in actual driving were permitted, and other restrictions or instructions were not provided. When acquiring actual driving data, as there was the risk of a traffic accident, rather than actually driving, a real vehicle (model name of SM5 new impression by Renault Samsung [41]) was started from a parked state in various locations (from daylight road to a parking garage).

In addition, we made two faster R-CNN models trained with our DDCD-DB1 and open database (CAVE-DB [1]), respectively, public.

 

1. Smith, B.A.; Yin, Q.; Feiner, S.K.; Nayar, S.K. Gaze Locking: Passive Eye Contact Detection for Human-Object Interaction. In Proceedings of the 26th Annual ACM Symposium on User Interface Software and Technology, St. Andrews, Scotland, UK, 8-11 October 2013; pp. 271280.

 

(2) Request for DDCD-DB1 and faster R-CNN model with algorithms

To gain access to DDCD-DB1 and faster R-CNN model with algorithms, download the following request form. Please scan the request form and email to Mr. Sung Ho Park (pshgod91@dongguk.edu).

Any work that uses this DDCD-DB1 and faster R-CNN model with algorithms must acknowledge the authors by including the following reference.

 

Sung Ho Park, Hyo Sik Yoon and Kang Ryoung Park, Faster R-CNN and Geometric Transformation-based Detection of Drivers Eyes Using Multiple NIR Camera Sensors, Sensors, in submission.

 

===========================================================================================================================================================================================================

 

< Request Form for DDCD-DB1 and faster R-CNN model with algorithms >

 

Please complete the following form to request access to the DDCD-DB1 and faster R-CNN model with algorithms (All contents must be completed). This database and CNN model with algorithms should not be used for commercial use.

 

Name :

 

Contact : (Email)

(Telephone)

 

Organization Name :

 

Organization Address :

 

Purpose :

 

 

Date :

 

               Name (signature)

 

 

 

26. Dongguk FRED-Net with Algorithm

 

(1) Introduction

We trained fully residual encoder-decoder network (FRED-Net) CNN models for iris and road scene segmentations, to evaluate the segmentation performance system using seven databases including NICE-II [1], MICHE [2], CASIA distance [3], CASIA interval [3], IITD [4], CamVid [5], and KITTI [6]. We made trained models/Algorithm open to other researchers.

 

6.     NICE.II. Noisy Iris Challenge Evaluation-Part II. Available online: http://nice2.di.ubi.pt/index.html (accessed on 28 December 2017).

7.     De Marsico, M.; Nappi, M.; Riccio, D.; Wechsler, H. Mobile iris challenge evaluation (MICHE)-I, biometric iris dataset and protocols. Pattern Recognit. Lett. 2015, 57, 1723.

8.     CASIA-Iris-databases. Available online: http://biometrics.idealtest.org/dbDetailForUser.do?id=4 (accessed on 28 December 2017).

9.     IIT Delhi Iris Database. Available online: http://www4.comp.polyu.edu.hk/~csajaykr/IITD/Database_Iris.htm (accessed on 28 December 2017).

10.   Brostow, G. J.; Fauqueur, J.; Cipolla, R. Semantic object classes in video: A high-definition ground truth database. Pattern Recognit. Lett. 2009, 30, 8897.

11.   Geiger, A.; Lenz, P.; Stiller, C.; Urtasun, R. Vision meets robotics: The KITTI dataset. Int. J. Robot. Res. 2013, 32, 12311237.

 

(2) FRED-Net Model with Algorithm Request

To gain access to the models and algorithm, download the following FRED-Net model with algorithm request form. Please sign and scan the request form and email to Mr. Arsalan (arsal@dongguk.edu).

Any work that uses this CNN model must acknowledge the authors by including the following reference.

 

Muhammad Arsalan, Dong Seop Kim, Min Beom Lee, Muhammad Owais, and Kang Ryoung Park, FRED-Net: Fully Residual Encoder-Decoder Network for Accurate Iris Segmentation, Expert Systems with Applications, in submission.

 

 

< FRED-Net model with algorithm Request Form >

 

Please complete the following form to request access to the FRED-Net model with algorithm (All contents must be completed). This CNN model with algorithm should not be used for commercial use.

 

Name:

 

Contact:  (Email)

(Telephone)

 

Organization Name:

 

Organization Address:

 

Purpose:

 

 

Date:

 

                Name (signature)

 

 

 

25. Dongguk Face and Body Database (DFB-DB1) with CNN models and algorithms

 

(1) Introduction

DFB-DB1 was created for the experiments using images of 22 people obtained by two types of cameras to assess the performance of the proposed method in a variety of camera environments. The first camera was a Logitech BCC 950, and the camera specifications include a camera viewing angle of 78°, a maximum resolution of full high-definition (HD) 1080 p, and auto-focusing at 30 frames per second (fps). The second camera was a Logitech C920, and its specifications include a maximum resolution of full HD 1080p, a viewing angle of 78° at 30 fps, and auto focusing. Images were taken in an indoor environment with indoor lights on, and each camera was installed at a height of 2 m 40 cm. The database was divided into two categories according to the camera. In the first database, the images were captured by the Logitech BCC 950, and the second database is composed of the images obtained by the Logitech C920, and the angle of camera was similar to that for capturing the first database.

In addition, we open our two CNN models trained with DFB-DB1 and open database of ChokePoint database [1], respectively, in addition to our algorithms.

 

1. ChokePoint Database. Available online: http://arma.sourceforge.net/chokepoint/ (accessed on 21 Feb. 2018).

 

(2) Request for DFB-DB1 with CNN model and algorithms

To gain access to DFB-DB1 with CNN model and algorithms, download the following request form. Please scan the request form and email to Mr. Ja Hyung Koo (koo6190@naver.com).

Any work that uses this DFB-DB1 with CNN model and algorithms must acknowledge the authors by including the following reference.

 

Ja Hyung Koo, Se Woon Cho, Na Rae Baek, Min Cheol Kim, and Kang Ryoung Park, CNN-Based Multimodal Human Recognition in Surveillance Environments,” Sensors, Vol. 18, Issue 9(3040), pp. 1-34, September 2018 

 

===========================================================================================================================================================================================================

 

< DFB-DB1 and CNN model Request Form >

 

Please complete the following form to request access to the DFB-DB1 and CNN model (All contents must be completed). This database and CNN model should not be used for commercial use.

 

Name :

 

Contact : (Email)

(Telephone)

 

Organization Name :

 

Organization Address :

 

Purpose :

 

 

Date :

 

               Name (signature)

 

 

 

24. Dongguk Night-Time Face Detection database (DNFD-DB1) and algorithm including CNN model

 

(1) Introduction

DNFD-DB1 is a self-constructed database acquired through a fixed single visible-light camera at a distance of approximately 2022 m at night. The resolution of the camera is 1600 × 1200 pixels, but the image is cropped to the average adult height, which is approximately 600. A total of 2,002 images of 20 different people were prepared, and there are 46 people in each frame. To carry out the 2-fold cross-validation, those 20 people were divided into two subsets of 10 people. In addition, we made two 2-stage Faster R-CNN models trained with our DNFD-DB1 and open database of Fudan University [1], respectively, public.

 

[1] Open database of Fudan University. Available online: https://cv.fudan.edu.cn/_upload/tpl/06/f4/1780/template1780/humandetection.htm (accessed on 26 March 2018).

 

(2) DNFD-DB1 and CNN model Request

To gain access to DNFD-DB1 and CNN model, download the following request form. Please scan the request form and email to Mr. Se Woon Cho (jsu319@naver.com).

Any work that uses this DNFD-DB1 and CNN model must acknowledge the authors by including the following reference.

 

Se Woon Cho, Na Rae Baek, Min Cheol Kim, Ja Hyung Koo, Jong Hyun Kim, and Kang Ryoung Park, Face Detection in Nighttime Images Using Visible-Light Camera Sensors with Two-Step Faster Region-Based Convolutional Neural Network, Sensors, Vol. 18, Issue 9(2995), pp. 1-31, September 2018.

 

===========================================================================================================================================================================================================

 

< DNFD-DB1 and CNN model Request Form >

 

Please complete the following form to request access to the DNFD-DB1 and CNN model (All contents must be completed). This database and CNN model should not be used for commercial use.

 

Name :

 

Contact : (Email)

(Telephone)

 

Organization Name :

 

Organization Address :

 

Purpose :

 

 

Date :

 

               Name (signature)

 

 

 

23. Dongguk Iris Spoof Detection CNN Model version 2 (DFSD-CNN-2) with Algorithm

 

(1) Introduction

We trained CNN models using local and global image features based on VGG-19-Net architecture for presentation attack detection for iris recognition system using two public databases, Warsaw-2017 [1] and Notre Dame2015 [2], respectively. We made trained models open to other researchers.

 

12.   Yambay, D.; Becker, B.; Kohli, N.; Yadav, D.; Czajka, A.; Bowyer, K. W.; Schuckers, S.; Singh, R.; Vatsa, M.; Noore, A.; Gragnaniello, D.; Sansone, C.; Verdoliva, L.; He, L.; Ru, Y.; Li, H.; Liu, N.; Sun, Z.; Tan, T. LivDet iris 2017 iris liveness detection competition 2017. In Proceedings of the International Conference on Biometrics, Denver, CO, USA, 1-4 October 2017.

13.   Doyle, J. S.; Bowyer, K. W. Robust detection of textured contact lens in iris recognition using BSIF. IEEE Access, 2015, 3, 1672-1683.

 

(2) DFSD-CNN-2 Model Request

To gain access to the models and algorithm, download the following DFSD-CNN-2 request form. Please sign and scan the request form and email to Prof. Nguyen (nguyentiendat@dongguk.edu).

 

Any work that uses this CNN model must acknowledge the authors by including the following reference.

 

D. T. Nguyen, T. D. Pham, Y. W. Lee, and K. R. Park, "Deep Learning-Based Enhanced Presentation Attack Detection for Iris Recognition by Combining Features from Local and Global Regions Based on NIR Camera Sensor," Sensors, Vol. 18, Issue 8(2601), pp. 1-32, August 2018.

 

===========================================================================================================================================================================================================

 

< DFSD-CNN-2 model Request Form >

 

Please complete the following form to request access to the DFSD-CNN-2 model (All contents must be completed). This CNN model should not be used for commercial use.

 

Name:

 

Contact:  (Email)

(Telephone)

 

Organization Name:

 

Organization Address:

 

Purpose:

 

 

Date:

 

                Name (signature)

 

 

 

22. Dongguk Fitness Database (DF-DB2) & CNN Model

 

(1) Introduction

 

We collected banknote fitness databases (DF-DB2) from three national currencies, which are Korean won (KRW), Indian rupee (INR), and Unites States dollar (USD). Six denominations exist in the INR dataset: 10, 20, 50, 100, 500, and 1000 rupees, and two denominations exist in the KRW dataset: 1000 and 5000 wons, each of which consists of three fitness levels of fit, normal, and unfit for recirculation, called the case 1 fitness level. In these case 1 datasets, each banknote image was captured using VR sensors on both sides, and IRT sensors on the front side. Five denominations exist for the USD: 5, 10, 20, 50 and 100 dollars, divided into two fitness levels of fit and unfit, called the case 2 fitness level. The number of images captured per banknote was two, including the VR and IRT images of one side of the banknote. In addition, we made CNN models trained with our DF-DB2 public.

 

 

(2) DF-DB2 and CNN model Request

To gain access to DF-DB2 and CNN models, download the following request form. Please scan the request form and email to Prof. Tuyen Danh Pham (phamdanhtuyen@gmail.com).

Any work that uses this DF-DB2 and CNN Model must acknowledge the authors by including the following reference.

 

Tuyen Danh Pham, Dat Tien Nguyen, Jin Kyu Kang, and Kang Ryoung Park, "Deep Learning-Based Multinational Banknote Fitness Classification with a Combination of Visible-Light Reflection and Infrared-Light Transmission Images," Symmetry-Basel, Vol. 10, Issue 10(431), pp. 1-26, October 2018.

 

===========================================================================================================================================================================================================

 

< DF-DB2 and CNN model Request Form >

 

Please complete the following form to request access to the DF-DB2 and CNN model (All contents must be completed). This database and CNN model should not be used for commercial use.

 

Name :

 

Contact : (Email)

(Telephone)

 

Organization Name :

 

Organization Address :

 

Purpose :

 

 

Date :

 

               Name (signature)

 

 

 

21. Dongguk-body-movement-based human identification database version 2 (DBMHI-DB2) & CNN Model

 

(1) Introduction

We have collected our database in both dark and bright environments. The database included both front and back view images of humans. Our database has been collected in five different places in different days with same camera heights. The database consists of data of 100 people including men and women. The database includes both thermal and visible light images but only thermal images have been utilized in this research. The people in our database have different heights and widths, and their sizes vary from 27 to 150 pixels in width and from 90 to 390 pixels in height. In addition, we made our trained CNN and CNN-LSTM model public.

 

(2) DBMHI-DB2 database & the trained CNN model Request

To gain access to the database and CNN model, download the following DBMHI-DB2 and CNN model request form. Please scan the request form and email to Mr. Ganbayar Batchuluun (ganabata87@dongguk.edu).

Any work that uses or incorporates the database and CNN model must acknowledge the authors by including the following reference.

 

Ganbayar Batchuluun, Hyo Sik Yoon, Jin Kyu Kang, and Kang Ryoung Park, "Gait-Based Human Identification by Combining Shallow Convolutional Neural Network-Stacked Long Short-Term Memory and Deep Convolutional Neural Network," IEEE Access, Vol. 6, pp. 63164-63186, October 2018.

 

===========================================================================================================================================================================================================

 

< DBMHI-DB2 database & the trained CNN model Request Form >

 

Please complete the following form to request access to the DBMHI-DB2 and CNN model (All contents must be completed). This database and CNN model should not be used for commercial use.

 

Name :

 

Contact : (Email)

(Telephone)

 

Organization Name :

 

Organization Address :

 

Purpose :

 

 

Date :

 

               Name (signature)

 

===========================================================================================================================================================================================================

 

 

20. Dongguk Multimodal Recognition CNN of Finger-vein and Finger shape (DMR-CNN) with Algorithm

 

(1) Introduction

We trained multimodal recognition system of finger-vein and finger shape based on ResNet-50 and ResNet-101 using two databases including the Shandong University homologous multi-modal traits (SDUMLA-HMT) [1] and the Hong Kong Polytechnic University Finger Image Database (version 1) [2]. We made trained models/Algorithm open to other researchers.

 

1. SDUMLA-HMT Finger Vein Database. Available online: http://mla.sdu.edu.cn/info/1006/1195.htm (accessed on 7 May 2018).

2. Kumar, A.; Zhou, Y. Human identification using finger images. IEEE Trans. Image Process. 2012, 21, 2228-2244.

 

(2) DMR-CNN Model with Algorithm Request

To gain access to the models and algorithm, download the following DMR-CNN model with algorithm request form. Please sign and scan the request form and email to Mr. Wan Kim (daiz0128@naver.com).

 

Any work that uses this CNN model must acknowledge the authors by including the following reference.

 

W. Kim, J. M. Song, and K. R. Park, Multimodal Biometric Recognition Based on Convolutional Neural Network by the Fusion of Finger-vein and Finger Shape Using Near-Infrared (NIR) Camera Sensor, Sensors, Vol. 18, Issue 7(2296), pp. 1-34, July 2018.

< DMR-CNN model with algorithm Request Form >

 

Please complete the following form to request access to the DMR-CNN model with algorithm (All contents must be completed). This CNN model with algorithm should not be used for commercial use.

 

Name:

 

Contact:  (Email)

(Telephone)

 

Organization Name:

 

Organization Address:

 

Purpose:

 

 

Date:

 

                Name (signature)

 

 

 

19. Dongguk Drone Camera Database (DDroneC-DB2) with CNN models

 

(1) Introduction

In our experiments, we used a DJI Phantom 4 quadcopter to capture the video while the drone was landing or hovering. It includes a color camera with a 1/2.3-inch-thick complementary metaloxidesemiconductor (CMOS) sensor, with a 94O-field-of-view (FOV) and an f/2.8 lens. The captured videos are in mpeg-4 (MP4) format with 30 fps, and have a size of 1280 x 720 pixels. The drones gimbal is adjusted 90° downward so that during landing, the camera can be facing the ground. In our database (shown in Table 1), we captured three videos, and acquired videos in varying types of environments (humidity level, wind velocity, temperature, and weather). We make our CNN model trained by this database and that trained by PASCAL VOC and Ms COCO databases open to other researchers, also.

 

Table 1. Description of Description of DDroneC-DB2

Sub-dataset

Number of images

Condition

Description

Morning

Far

3088

Humidity: 44.7%

Wind speed: 5.2 m/s

Temperature: 15.2 °C, autumn,sunny

Illuminance:1800 lux

Landing speed: 5.5 m/s

Auto mode of camera shutter speed (8~1/8000 s) and ISO (100~3200)

Close

641

Close

(from DdroneC-DB1)

425

Humidity: 41.5 %

Wind speed: 1.4 m/s

Temperature: 8.6 °C,

spring, sunny

Illuminance: 1900 lux

Landing speed: 4 m/s

Auto mode of camera shutter speed

(8~1/8000 s) and ISO (100~3200)

Afternoon

Far

2140

Humidity: 82.1%

Wind speed: 6.5 m/s

Temperature: 28 °C, summer, sunny

Illuminance:2250 lux

Landing speed: 7 m/s

Auto mode of camera shutter speed (8~1/8000 s) and ISO (100~3200)

Close

352

Close

(from DdroneC-DB1)

148

Humidity: 73.8 %

Wind speed: 2 m/s

Temperature: -2.5 °C,

winter, cloudy

Illuminance: 1200 lux

Landing speed: 6 m/s

Auto mode of camera shutter speed

(8~1/8000 s) and ISO (100~3200)

Evening

Far

3238

Humidity: 31.5%

Wind speed: 7.2 m/s

Temperature: 6.9 °C, autumn, foggy

Illuminance: 650 lux

Landing speed: 6 m/s

Auto mode of camera shutter speed (8~1/8000 s) and ISO (100~3200)

Close

326

Close

(from DdroneC-DB1)

284

Humidity: 38.4 %

Wind speed: 3.5 m/s

Temperature: 3.5 °C,

winter, windy

Illuminance: 500 lux

Landing speed: 4 m/s

Auto mode of camera shutter speed

(8~1/8000 s) and ISO (100~3200)

 

 

(2) Request for DDroneC-DB2 and CNN models

To gain access to the database with CNN models, download the following request form. Please scan the request form and email to Mr. Phong Ha Nguyen (stormwindvn@dongguk.edu).

Any work that uses or incorporates the database must acknowledge the authors by including the following reference.

 

Phong Ha Nguyen, Muhammad Arsalan, Ja Hyung Koo, Rizwan Ali Naqvi, Noi Quang Truong, and Kang Ryoung Park, LightDenseYOLO: A Fast and Accurate Marker Tracker for Autonomous UAV Landing by Visible Light Camera Sensor on Drone, Sensors, Vol. 18, Issue 6(1703), pp. 1-30, May 2018.

 

===========================================================================================================================================================================================================

 

< Request Form for DDroneC-DB2 and CNN models >

 

Please complete the following form to request access to the DDroneC-DB2 and CNN models (All contents must be completed). This database should not be used for commercial use.

 

Name :

 

Contact : (Email)

(Telephone)

 

Organization Name :

 

Organization Address :

 

Purpose :

 

 

Date :

 

               Name (signature)

 

 

 

18. Dongguk Periocular Database (DP-DB1) with CNN models and algorithms

 

(1) Introduction

The DP-DB1 database was created for research on periocular recognition in an indoor surveillance environment. The camera used to capture the images was a Logitech BCC 950, and the specifications of the camera include a camera viewing angle of 79 degrees, a maximum resolution of full high definition (Full HD) 1080p, and a frame rate of 30 fps with auto focusing. The location where the images were captured was an indoor hallway (with indoor lights on), and the camera was installed at a height of 2 m 40 cm. This database consists of 20 people captured in three scenarios: straight line movement, corner movement, and standing still. In case of the standing still scenario, the images were acquired from 4 different positions. In addition, we open our two CNN models trained with DP-DB1 and open database of ChokePoint database [1, 2], respectively, in addition to our algorithms.

 

1. Wong, Y.; Chen, S.; Mau, S.; Sanderson, C.; Lovell, B. C. Patch-based probabilistic image quality assessment for face selection and improved video-based face recognition. In Proceedings of IEEE Conference on Computer Vision and Pattern Recognition Workshops, Colorado Springs, CO, USA, 20-25 June 2011; pp. 74-81.

2. ChokePoint Database. Available online: http://arma.sourceforge.net/chokepoint/ (accessed on 21 Feb. 2018).

 

(2) Request for DP-DB1 with CNN model and algorithms

To gain access to DP-DB1 with CNN model and algorithms, download the following request form. Please scan the request form and email to Mr. Min Cheol Kim (mincheol9166@naver.com).

Any work that uses this DP-DB1 with CNN model and algorithms must acknowledge the authors by including the following reference.

 

Min Cheol Kim, Ja Hyung Koo, Se Woon Cho, Na Rae Baek, and Kang Ryoung Park, Convolutional Neural Network-based Periocular Recognition in Surveillance Environments, IEEE Access, Vol. 6, pp. 57291  57310, October 2018

 

===========================================================================================================================================================================================================

 

< Request form for DP-DB1 with CNN model and algorithms >

 

Please complete the following form to request access to the DP-DB1 with CNN model and algorithms (All contents must be completed). This database and CNN model with algorithms should not be used for commercial use.

 

Name :

 

Contact : (Email)

(Telephone)

 

Organization Name :

 

Organization Address :

 

Purpose :

 

 

Date :

 

               Name (signature)

 

 

 

17. Dongguk IrisDenseNet CNN Model (DI-CNN) with Algorithm

 

(1) Introduction

We trained IrisDenseNet CNN models based on DenseNet and SegNet architecture for iris segmentation, to evaluate the segmentation performance system using five databases including NICE-II [1], MICHE [2], CASIA distance [3], CASIA interval [3] and IITD [4]. We made trained models/Algorithm open to other researchers.

 

14.  NICE.II. Noisy Iris Challenge Evaluation-Part II. Available online: http://nice2.di.ubi.pt/index.html (accessed on 28 December 2017).

15.   De Marsico, M.; Nappi, M.; Riccio, D.; Wechsler, H. Mobile iris challenge evaluation (MICHE)-I, biometric iris dataset and protocols. Pattern Recognit. Lett. 2015, 57, 1723.

16.   CASIA-Iris-databases. Available online: http://biometrics.idealtest.org/dbDetailForUser.do?id=4 (accessed on 28 December 2017).

17.   IIT Delhi Iris Database. Available online: http://www4.comp.polyu.edu.hk/~csajaykr/IITD/Database_Iris.htm (accessed on 28 December 2017).

 

(2) DI-CNN Model with Algorithm Request

To gain access to the models and algorithm, download the following DI-CNN model with algorithm request form. Please sign and scan the request form and email to Mr. Arsalan (arsal@dongguk.edu).

 

Any work that uses this CNN model must acknowledge the authors by including the following reference.

 

Muhammad Arsalan, Rizwan Ali Naqvi, Dong Seop Kim, Phong Ha Nguyen, Muhammad Owais and Kang Ryoung Park, IrisDenseNet: Robust Iris Segmentation Using Densely Connected Fully Convolutional Networks in the Images by Visible Light and Near-Infrared Light Camera Sensors, Sensors, Vol. 18, Issue 5(1501), pp. 1-30, May 2018.

< DI-CNN model with algorithm Request Form >

 

Please complete the following form to request access to the DI-CNN model with algorithm (All contents must be completed). This CNN model with algorithm should not be used for commercial use.

 

Name:

 

Contact:  (Email)

(Telephone)

 

Organization Name:

 

Organization Address:

 

Purpose:

 

 

Date:

 

                Name (signature)

 

 

 

16. Dongguk Visible Light Iris Recognition CNN Model (DVLIR-CNN)

 

(1) Introduction

We made the recognition algorithm of iris region by two three convolutional neural networks (CNNs) trained with NICE-II training database [1, 2], the mobile iris challenge evaluation (MICHE) data [3, 4], and CASIA-Iris-Distance database [5], respectively. We made these trained CNN models open to other researchers.

 

18.   NICE.II. Noisy Iris Challenge Evaluation-Part II. Available online: http://nice2.di.ubi.pt/index.html (accessed on 26 July 2017).

19.   Proença, H.; Filipe, S.; Santos, R.; Oliveira, J.; Alexandre, L. A. The UBIRIS.v2: A database of visible wavelength iris images captured on-the-move and at-a-distance.  IEEE Trans. Pattern Anal. Mach. Intell. 2010, 32, 1529-1535.

20.   de Marsico, M.; Nappi, M.; Ricco, D.; Wechsler, H. Mobile iris challenge evaluation (MICHE)-I, biometric iris dataset and protocols. Pattern Recognit. Lett. 2015, 57, 17-23.

21.   Haindl, M.; Krupička, M. Unsupervised detection of non-iris occlusions. Pattern Recognit. Lett. 2015, 57, 60-65.

22.   CASIA-Iris-Distance. Available online: http://biometrics.idealtest.org/dbDetailForUser.do?id=4 (accessed on 13 November 2017).

 

(2) DVLIR-CNN model Request

To gain access to the CNN models, download the following DVLIR-CNN model request form. Please scan the request form and email to Mr. Min Beom Lee (mblee@dongguk.edu).

Any work that uses this CNN Model must acknowledge the authors by including the following reference.

 

Min Beom Lee, Hyung Gil Hong, and Kang Ryoung Park, "Noisy Ocular Recognition Based on Three Convolutional Neural Networks," Sensors, Vol. 17, Issue 12(2933), pp. 1-26, December 2017

 

===========================================================================================================================================================================================================

 

< DVLIR-CNN model Request Form >

 

Please complete the following form to request access to the DVLIR-CNN model (All contents must be completed). This CNN model should not be used for commercial use.

 

Name :

 

Contact : (Email)

(Telephone)

 

Organization Name :

 

Organization Address :

 

Purpose :

 

 

Date :

 

               Name (signature)

 

 

 

 

15. Dongguk Aggressive and Smooth Driving Database (DASD-DB1) and CNN Model

 

(1) Introduction

We used 15 subjects in the experiment. All the subjects voluntarily participated in our experiments. Because it was too risky to create an aggressive driving situation under real traffic conditions, we utilized two types of driving simulator, to assess baseline aggressive and smooth driving situations. As illustrated in Figure 1, the experiment included 5 min of smooth driving and another 5 min of aggressive driving. Between each section of the experiment, every subject watched a sequence of neutral images from the international affective picture system, thereby maintaining neutral emotional input. After the experiment, the subjects rested for about 10 min. This procedure was repeated three times.

 

(2) DASD-DB1 and CNN model Request

To gain access to DASD-DB1 and CNN models, download the following request form. Please scan the request form and email to Mr. Kwan Woo Lee (leekwanwoo@dgu.edu).

Any work that uses this DASD-DB1 and CNN Model must acknowledge the authors by including the following reference.

 

Kwan Woo Lee, Hyo Sik Yoon, Jong Min Song, and Kang Ryoung Park, Convolutional Neural Network-Based Classification of Drivers Emotion during Aggressive and Smooth Driving Using Multi-Modal Camera Sensors,Sensors, Vol. 18, Issue 4(957), pp. 1-22, March 2018

 

===========================================================================================================================================================================================================

 

< DASD-DB1 and CNN model Request Form >

 

Please complete the following form to request access to the DASD-DB1 and CNN model (All contents must be completed). This database and CNN model should not be used for commercial use.

 

Name :

 

Contact : (Email)

(Telephone)

 

Organization Name :

 

Organization Address :

 

Purpose :

 

 

Date :

 

               Name (signature)

 

 

 

 

14. Dongguk Night-time Pedestrian Detection Faster R-CNN and Algorithm

 

(1) Introduction

 

We made modified faster R-CNN model with algorithm for pedestrian detection at nighttime with the augmented images from KAIST database [1] and Caltech database [2]. We made this trained CNN model open to other researchers.

 

1. Hwang, S.; Park, J.; Kim, N.; Choi, Y.; Kweon, I.S. Multispectral pedestrian detection: Benchmark dataset and baseline. In Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, Boston, MA, USA, 7-12 June 2015; pp. 1037-1045.

2. Dollár, P.; Wojek, C.; Schiele, B.; Perona, P. Pedestrian detection: An evaluation of the state of the art. IEEE Trans. Pattern Anal. Mach. Intell. 2012, 34, 743-761.

 

(2) Modified faster R-CNN model with algorithm Request

To gain access to the CNN models with algorithm, download the following request form. Please scan the request form and email to Mr. Jong Hyun Kim (zzingae@dongguk.edu).

Any work that uses this CNN Model with algorithm must acknowledge the authors by including the following reference.

 

Jong Hyun Kim, Ganbayar Batchuluun, and Kang Ryoung Park, Pedestrian Detection Based on Faster R-CNN in Nighttime by Fusing Deep Convolutional Features of Successive Images, Expert Systems with Applications, (in press).

 

===========================================================================================================================================================================================================

 

< Modified faster R-CNN model with algorithm Request Form >

 

Please complete the following form to request access to this CNN model with algorithm (All contents must be completed). This CNN model with algorithm should not be used for commercial use.

 

Name :

 

Contact : (Email)

(Telephone)

 

Organization Name :

 

Organization Address :

 

Purpose :

 

 

Date :

 

               Name (signature)

 

 

 

 

13. Dongguk Shadow Detection Database (DSDD-DB1) & CNN Model

 

(1) Introduction

 

DSDD-DB1 was obtained by installing visual light cameras 5 to 10 m above the ground, which approximates the conventional height of surveillance camera. As shown in Figure 1 and Table 1, images are shot in the morning, the afternoon, the evening, and on rainy days under various weather conditions, temperature, and illumination. A total of 24,000 images, constituting five sub-datasets, are obtained. The original image size is 800 x 600 pixels of the RGB three-channel.

 

Table 1. Description of five datasets.

Dataset

Condition

Detail Description

I

0.9, afternoon, sunny

humidity 24 %, wind 3.6 m/s

- Shadow with dark color cast due to strong sunlight.

II

6.0, afternoon, cloudy, humidity 39 %, wind 1.9 m/s

- Sunlight weakened by cloud so that a shadow of lighter color is cast.

III

8.0, evening, cloudy, humidity 42 %, wind 3.5 m/s

- Darker image due to weak evening sunlight.

- Long and many shadows due to the sun position in the evening and the reflection on buildings.

IV

5.2, morning, sunny humidity 37 %, wind 0.6 m/s

- Background and object become less distinguishable due to strong morning sunlight.

V

13.8, afternoon, rainy, humidity 65 %, wind 2.0 m/s

- Overall dark image due to rainy day.

- Many shadows generated by wet background floor.

 

In addition, we made two CNN models trained with our DSDD-DB1 and open database (CAVIAR [1]), respectively, public.

 

[1] CAVIAR: Context Aware Vision using Image-based Active Recognition. Available online: http://homepages.inf.ed.ac.uk/rbf/CAVIAR/ (accessed on 8 August 2017).

 

(2) DSDD-DB1 and CNN model Request

To gain access to DSDD-DB1 and CNN models, download the following request form. Please scan the request form and email to Mr. Dong Seop Kim (k_ds1028@naver.com).

Any work that uses this DSDD-DB1 and CNN Model must acknowledge the authors by including the following reference.

 

Dong Seop Kim, Muhammad Arsalan, and Kang Ryoung Park, Convolutional Neural Network-Based Shadow Detection in Images Using Visible Light Camera Sensor, Sensors, Vol. 18, Issue 4(960), pp. 1-19, March 2018.

 

===========================================================================================================================================================================================================

 

< DSDD-DB1 and CNN model Request Form >

 

Please complete the following form to request access to the DSDD-DB1 and CNN model (All contents must be completed). This database and CNN model should not be used for commercial use.

 

Name :

 

Contact : (Email)

(Telephone)

 

Organization Name :

 

Organization Address :

 

Purpose :

 

 

Date :

 

               Name (signature)

 

 

 

12. Dongguk driver gaze classification database (DDGC-DB1) and CNN model

 

(1) Introduction

 

17 spots (gaze zones) were designated to gaze at for the experiment, and each driver stared at each spot five times. Data were collected from 20 drivers including 3 wearing glasses. The image size is 1600´1200 pixels with 3 channels. When the participants were staring at each spot, they were told to act normally, as if they were actually driving and were not restrained to one position or given any special instructions to act in an unnatural manner. There were risks of car accidents to motivate the participants to accurately stare at the 17 designated spots while actually driving for the experiment. Instead, this study obtained images from various locations (from roads in daylight to a parking garage) in a real vehicle (model name of SM5 New Impression by Renault Samsung) with its power on, but in park to create an environment most similar to when it is being driven (including factors like car vibration and external light). Moreover, to understand the influence of various kinds of external light on driver gaze detection, test data were acquired at different times of the day: in the morning, the afternoon, and at night.

In addition, we made two CNN models trained with our DDGC-DB1 and open database (CAVE-DB [1]), respectively, public.

 

[1] Smith, B.A.; Yin, Q.; Feiner, S.K.; Nayar, S.K. Gaze Locking: Passive Eye Contact Detection for Human-Object Interaction. In Proceedings of the 26th Annual ACM Symposium on User Interface Software and Technology, St. Andrews, Scotland, UK, 8-11 October 2013; pp. 271280.

 

(2) DDGC-DB1 and CNN model Request

To gain access to DDGC-DB1 and CNN models, download the following request form. Please scan the request form and email to Mr. Rizwan Ali Naqvi (rizwanali@dongguk.edu).

Any work that uses this DDGC-DB1 and CNN Model must acknowledge the authors by including the following reference.

 

Rizwan Ali Naqvi, Muhammad Arsalan, Ganbayar Batchuluun, Hyo Sik Yoon, and Kang Ryoung Park, Deep Learning-Based Gaze Detection System for Automobile Drivers Using a NIR Camera Sensor, Sensors, Vol. 18, Issue 2(456), pp. 1-34, February 2018.

 

===========================================================================================================================================================================================================

 

< DDGC-DB1 and CNN model Request Form >

 

Please complete the following form to request access to the DDGC-DB1 and CNN model (All contents must be completed). This database and CNN model should not be used for commercial use.

 

Name :

 

Contact : (Email)

(Telephone)

 

Organization Name :

 

Organization Address :

 

Purpose :

 

 

Date :

 

               Name (signature)

 

 

 

 

11. Dongguk Age Estimation CNN Model (DAE-CNN)

 

(1) Introduction

 

We made age estimation robust to optical and motion blurring by Resnet-152 trained with PAL database including artificially (optical and motion) blurred images [1, 2] and MORPH database including artificially (optical and motion) blurred images [3], respectively. We made these trained CNN model open to other researchers.

 

23.   Minear, M.; Park, D.C. A lifespan database of adult facial stimuli. Behav. Res. Methods Instrum. Comput. 2004, 36, 630633.

24.  PAL database. Available online: http://agingmind.utdallas.edu/download-stimuli/face-database/ (accessed on 17 May 2017).

25.   MORPH database. Available online: https://ebill.uncw.edu/C20231_ustores/web/store_main.jsp?STOREID=4 (accessed on 17 May 2017).

 

(2) DAE-CNN model Request

To gain access to the CNN models, download the following DAE-CNN model request form. Please scan the request form and email to Mr. Jeon Seong Kang (kjs2605@dgu.edu).

Any work that uses this CNN Model must acknowledge the authors by including the following reference.

 

Jeon Seong Kang, Chan Sik Kim, Young Won Lee, Se Woon Cho, and Kang Ryoung Park, Age Estimation Robust to Optical and Motion Blurring by Deep Residual CNN, Symmetry-Basel, Vol. 10, Issue 4(108), pp. 1-23, April 2018. 

 

===========================================================================================================================================================================================================

 

< DAE-CNN model Request Form >

 

Please complete the following form to request access to the DAE-CNN model (All contents must be completed). This CNN model should not be used for commercial use.

 

Name :

 

Contact : (Email)

(Telephone)

 

Organization Name :

 

Organization Address :

 

Purpose :

 

 

Date :

 

               Name (signature)

 

 

 

 

10. Dongguk Single Camera-based Driver Database (DSCD-DB1)

 

(1) Introduction

 

We collected DSCD-DB1 from a total of 26 participants: 10 wearing nothing, 8 wearing only four kinds of glasses, 5 wearing only two kinds of sunglasses, and 3 wearing only hat. In addition, even the people (wearing nothing) took various pose including putting one hand to cheek or using mobile phone. Fifteen spots in car were designated to gaze at for the experiment, and each participant stared at each spot five times. When the participants were staring at each spot, they were told to act normally, as if they were actually driving and were not restrained to one position or given any special instructions to act in an unnatural manner. There were risks of car accidents and such to motivate the participants to accurately stare at the 15 designated spots while actually driving for the experiment. Instead, this study obtained images from various locations (from roads in daylight to a parking garage) in a real vehicle (model name of SM5 New Impression by Renault Samsung) with its power on, but in park in order to create an environment most similar to when it is being driven (including factors like car vibration and external light). Moreover, to understand the influence of various kinds of external light on driver gaze detection, test data were acquired at different times of the day: in the morning, the afternoon, and at night.

In addition, we collected the data from the participants (sitting at the side seat of driver) gazing at 15 spots while the driver was actually driving the car. Because the data were collected from the participant sitting at the side seat of driver by attaching our gaze tracking device in front of the participant, the conditions of data acquisition were similar to those from driver. In order to check our method in various car environments, we used different car (model name of Daewoo Lacetti Premiere by Chevrolet). The database was collected from a total of 10 participants: 3 wearing nothing, 3 wearing only three kinds of glasses, 2 wearing only two kinds of sunglasses, and 2 wearing only hat. In addition, even the people (wearing nothing) took various pose including putting one hand to cheek or using mobile phone. When the participants were staring at each spot, they were told to act normally, as if they were actually driving and were not restrained to one position or given any special instructions to act in an unnatural manner. To understand the influence of various kinds of external light on driver gaze detection, test data were acquired at different times of the day: in the morning, the afternoon, and at night, and they were collected while driving on various roads.

Data were obtained and processed on a laptop computer with 2.80 GHz CPU (Intel ® Core i5-4200H) and 8 GB of RAM.

 

(2) DSCD-DB1 database Request

To gain access to the database, download the following DSCD-DB1 request form. Please scan the request form and email to Mr. Hyo Sik Yoon (yoonhs@dongguk.edu).

Any work that uses or incorporates the database must acknowledge the authors by including the following reference.

 

Hyo Sik Yoon, Hyung Gil Hong, Dong Eun Lee, and Kang Ryoung Park, Driver's Eye-based Gaze Tracking System by One-Point Calibration, Multimedia Tools and Applications, 2019, in press.

 

===========================================================================================================================================================================================================

 

< DSCD-DB1 database Request Form >

 

Please complete the following form to request access to the DSCD-DB1 (All contents must be completed). This database should not be used for commercial use.

 

Name :

 

Contact : (Email)

(Telephone)

 

Organization Name :

 

Organization Address :

 

Purpose :

 

 

Date :

 

               Name (signature)

 

===========================================================================================================================================================================================================

 

 

9. ISPR Database (real and presentation attack finger-vein images) & Algorithm Including CNN Model

 

(1) Introduction

The ISPR database consists of 3300 and 7560 images for real and presentation attack finger-vein images, respectively. The real finger-vein database was collected by capturing finger-vein images from 33 people, including both male and female. For each person, all 10 fingers were used and 10 trials were captured for each finger. Consequently, the real finger-vein database contains 3300 (33 people × 10 fingers × 10 trials) real finger-vein images for our experiments. Among 3300 real finger-vein images, we selected 56 fingers which have clear vein pattern for making the presentation attack finger-vein images.

There are two reasons for this selection scheme. First, the users should normally use a finger with clear vein pattern for finger-vein recognition system in order to guarantee the security level of biometric feature. Therefore, if an attacker steals a finger-vein pattern image from user, it will normally be a clear finger-vein image. Second, with some fingers which normally have poor vein pattern such as thumbs or little fingers it is very hard for attackers to produce a clear presentation attack finger-vein image. Instead, the presentation attack finger-vein images can contain poor vein pattern and much of noise due to the reproducing process. As a result, the attacking task can be failed by rejection rate of finger-vein recognition system.

The presentation attack finger-vein image database was collected by re-capturing the printed versions of 56 selected real finger-vein images in three different printing materials including A4 paper, MAT paper and OHP film. In addition, we used three different values of the printing resolution of low resolution (300dpi), middle resolution (1200dpi) and high resolution (2400dpi). By using this scheme, we can collect presentation attack finger-vein images which contain various characteristics about printing materials and printing resolution. Finally, in order to simulate the attacking process, we captured presentation attack finger-vein images at three z-distances (the distance between camera and finger-vein sample) by little changing the z-distance during image acquisition and 5 trials for each z-distance.

As a result, a presentation attack finger-vein image database of 7560 images (56 real image × 3 printing materials × 3 printing resolutions × 3 z-distances × 5 trials) was collected.

 

Table 1. Description of ISPR presentation attack finger-vein image database

Image Making Protocol

Real Access

Printed Access

Train Set

Test Set

Total

Train Set

Test Set

Total

Material

Printed on A4 Paper

(ISPR-DB1)

1700

1600

3300

1440

1080

2520

Printed on MAT Paper

(ISPR-DB2)

1700

1600

3300

1440

1080

2520

Printed on OHP Film

(ISPR-DB3)

1700

1600

3300

1440

1080

2520

Printer Resolution

Printed using 300 DPI Resolution Printer

(ISPR-DB4)

1700

1600

3300

1440

1080

2520

Printed using 1200 DPI Resolution Printer. (ISPR-DB5)

1700

1600

3300

1440

1080

2520

Printed using 2400 DPI Resolution Printer

(ISPR-DB6)

1700

1600

3300

1440

1080

2520

Entire Database (ISPR-DB)

1700

1600

3300

4320

3120

7560

 

In addition, we made our algorithm including the trained CNN model public.

 

(2) ISPR database & algorithm including the trained CNN model Request

To gain access to the database and algorithm, download the following ISPR database and algorithm request form. Please scan the request form and email to Prof. Dat Tien Nguyen (nguyentiendat@dongguk.edu).

Any work that uses or incorporates the database and algorithm must acknowledge the authors by including the following reference.

 

Dat Tien Nguyen, Hyo Sik Yoon, Tuyen Danh Pham, and Kang Ryoung Park, "Spoof Detection for Finger-Vein Recognition System Using NIR Camera," Sensors, Vol. 17, Issue 10(2261), pp. 1-33, October 2017.

 

===========================================================================================================================================================================================================

 

< ISPR Database & Algorithm Request Form >

 

Please complete the following form to request access to the ISPR database and algorithm (All contents must be completed). This database and algorithm should not be used for commercial use.

 

Name :

 

Contact : (Email)

(Telephone)

 

Organization Name :

 

Organization Address :

 

Purpose :

 

 

Date :

 

               Name (signature)

 

===========================================================================================================================================================================================================

 

 

 

8. Dongguk Visible Light & FIR Pedestrian Detection Database (DVLFPD-DB1) && CNN model

 

(1) Introduction

 

There are 4 sub-databases, and the total number of frames of visible light images and FIR images is 4080 each. To obtain the images, this study used a duel camera system consisting of a FLIR Tau640 FIR camera (19 mm), and a Logitech C600 visible light web-camera. In order to record the filming conditions, a WH-1091 (wireless weather station) was used.

 

Table 1. Description of Database

 

Sub-database

1

Sub-database

2

Sub-database

3

Sub-database

4

Number of

image

598

651

2364

467

Number of      pedestrian candidate

1123

566

2432

169

Number of           non-pedestrian candidate

763

734

784

347

(range of width) ´ (range of height)

(pixels)

Pedestrian

(27 ~ 91) ´     (87 ~ 231)

(47 ~ 85) ´   (85 ~ 163)

(31 ~ 105) ´  (79 ~ 245)

(30 ~ 40) ´   (90 ~ 120)

Non-pedestrian

(51 ~ 161) ´   (63 ~ 142)

(29 ~ 93) ´   (49 ~ 143)

(53 ~ 83) ´  (55 ~ 147)

(60 ~ 170) ´  (50 ~ 110)

Weather Conditions

Surface Temperature: 30.4°C,

Air Temperature: 22.5°C  

Wind speed: 10 km/h   Sensory Temperature: 21.3°C

Surface Temperature: 25.5°C

Air Temperature: 24 Wind speed: 5 km/h Sensory Temperature: 23.5°C

Surface Temperature: 20°C

Air Temperature: 21°C    

Wind Speed: 6.1 km/h   Sensory Temperature: 21°C

Surface Temperature: 16°C

Air Temperature: 20.5°C   

Wind Speed: 2.5 km/h   Sensory Temperature: 20.8°C

 

In addition, we obtained the artificially degraded database by Gaussian noise and Gaussian blurring with the original dataset of Table 1, and make this degraded dataset public, also.

 

Also, we made the classification algorithm of pedestrian and non-pedestrian by convolutional neural network (CNN) trained with DVLFPD-DB1. We also made this trained CNN model open to other researchers.

 

 

(2) DVLFPD-DB1 & CNN model Request

To gain access to the database and CNN models, download the following DVLFPD-DB1 & CNN model request form. Please scan the request form and email to Mr. Jin Kyu Kang (kangjinkyu@dgu.edu).

Any work that uses or incorporates the database & CNN Model must acknowledge the authors by including the following reference.

 

Jin Kyu Kang, Hyung Gil Hong, and Kang Ryoung Park, Pedestrian Detection Based on Adaptive Selection of Visible Light or Far-Infrared Light Camera Image by Fuzzy Inference System and Convolutional Neural Network-Based Verification, Sensors, Vol. 17, Issue 7(1598), pp. 1-32, July 2017

 

===========================================================================================================================================================================================================

 

< DVLFPD-DB1 & CNN model Request Form >

 

Please complete the following form to request access to the DVLFPD-DB1 & CNN model (All contents must be completed). This database and CNN model should not be used for commercial use.

 

Name :

 

Contact : (Email)

(Telephone)

 

Organization Name :

 

Organization Address :

 

Purpose :

 

 

Date :

 

               Name (signature)

 

 

 

 

7. Dongguk Open and Closed Eyes Database (DOCE-DB1) & CNN model

 

(1) Introduction

 

DOCE-DB1 was created in an environment with a person watching TV indoors, with images taken at distances between 2 and 2.5 m. The camera used for capturing the images was a Logitech C600 web camera with a zoom lens attached. We used eye images obtained by extracting facial and eye features from images captured at a resolution of 1600 × 1200 pixels. Table 1 shows the descriptions of original and augmented DOCE-DB1.

 

Table 1. Descriptions of original and augmented databases

 

Original database

Augmented database

Open eye images

2,062

103,100

Closed eye images

4,763

238,150

Total

6,825

341,250

 

In addition, we made the classification algorithm of open and closed eye by convolutional neural network (CNN) trained with DOCE-DB1. We also made this trained CNN model with other various (trained) CNN models (based on AlexNet, VGG face, GoogLeNet) open to other researchers.

 

 

(2) DOCE-DB1 & CNN model Request

To gain access to the database and CNN models, download the following DOCE-DB1 & CNN model request form. Please scan the request form and email to Mr. Ki Wan Kim (yawara18@dgu.edu).

Any work that uses or incorporates the database & CNN Model must acknowledge the authors by including the following reference.

 

Ki Wan Kim, Hyung Gil Hong, Gi Pyo Nam, and Kang Ryoung Park, A Study of Deep CNN-Based Classification of Open and Closed Eyes Using a Visible Light Camera Sensor, Sensors, Vol. 17, Issue 7(1534), pp. 1-21, July 2017

 

===========================================================================================================================================================================================================

 

< DOCE-DB1 & CNN model Request Form >

 

Please complete the following form to request access to the DOCE-DB1 & CNN model (All contents must be completed). This database and CNN model should not be used for commercial use.

 

Name :

 

Contact : (Email)

(Telephone)

 

Organization Name :

 

Organization Address :

 

Purpose :

 

 

Date :

 

               Name (signature)

 

 

 

 

6. Dongguk Multi-national Currencies Database (DMC-DB1) & CNN model

 

(1) Introduction

Our multi-national banknote image database (DMC-DB1) containing images from six national currencies, which are CNY, EUR, JPY, KRW, RUB and USD. There were 64,668 images captured in both sides of 32,334 banknotes belonging to 62 denominations of currencies from six countries. The number of classes is four times of the number of denominations because of the inclusion of four directions; therefore, there are totally 248 classes of banknote to be classified in our study. In Table 1, we give the details numbers of images and classes (denominations and directions) of each countrys banknote in the dataset.

 

Table 1. Number of images and classes in our multi-national banknote database

Currency

Number of Images

Number of Classes

CNY

626

40

EUR

4,324

44

JPY

1,462

28

KRW

536

28

RUB

12,146

40

USD

45,574

68

 

In addition, we made the classification algorithm of multi-national currencies by convolutional neural network (CNN) trained with DMC-DB1. We also made this trained CNN model open to other researchers.

 

 

(2) DMC-DB1 & CNN model Request

To gain access to the database and CNN model, download the following DMC-DB1 & CNN model request form. Please scan the request form and email to Prof. Tuyen Danh Pham (phamdanhtuyen@gmail.com).

Any work that uses or incorporates the database & CNN Model must acknowledge the authors by including the following reference.

 

Tuyen Danh Pham, Dong Eun Lee, and Kang Ryoung Park, "Multi-National Banknote Classification Based on Visible-light Line Sensor and Convolutional Neural Network," Sensors, Vol. 17, Issue 7(1595), pp. 1-20, July 2017

 

===========================================================================================================================================================================================================

 

< DMC-DB1 & CNN model Request Form >

 

Please complete the following form to request access to the DMC-DB1 & CNN model (All contents must be completed). This database and CNN model should not be used for commercial use.

 

Name :

 

Contact : (Email)

(Telephone)

 

Organization Name :

 

Organization Address :

 

Purpose :

 

 

Date :

 

               Name (signature)

 

 

 

 

5. Dongguk Finger-Vein Database (DFingerVein-DB1) & CNN Model

 

(1) Introduction

DFingerVein-DB1 includes two databases. The first database (good-quality database) was acquired with two guide bars (fingertip guide bar and finger side guide bar) attached, when the images were captured using the finger-vein capturing device made by our lab. Ten images each were captured from the index, middle, and ring fingers of the left and right hands of 20 persons. Thus, this database has a total of 1,200 images consisting of 20 people × 2 hands × 3 fingers × 10 images. This is called good-quality database, because the images have few finger-vein misalignments by finger translation owing to the two guide bars, and there is little shading in the finger-vein images.

For the second database (mid-quality database), the two guide bars were removed when the images were captured with the same device used for the collection of good-quality database. Ten images each were captured from the index, middle, and ring fingers of the left and right hands of 33 persons. Thus, this database has a total of 1,980 images consisting of 33 people × 2 hands × 3 fingers × 10 images. This is called mid-quality database because the images were captured with no guide bar, and the images have finger-vein misalignments by finger translation; however, there is little shading in the finger-vein images similar to the good-quality database.

 

Table 1. Descriptions of the collected databases

 

Good-quality database

Mid-quality database

Original images

# of images

1,200

1,980

# of people

20

33

# of hands

2

2

# of fingers*

3

3

# of classes

(# of images per class)

120

(10)

198

(10)

 

In addition, we made the finger-vein recognition algorithm by convolutional neural network (CNN) trained with DFingerVein-DB1. We also made this trained CNN model open to other researchers.

 

 

(2) DFingerVein-DB1 & CNN Model Request

To gain access to the database & CNN Model, download the following DFingerVein-DB1 & CNN Model request form. Please scan the request form and email to Mr. Hyung Gil Hong (hell@dongguk.edu).

Any work that uses or incorporates the database & CNN Model must acknowledge the authors by including the following reference.

 

Hyung Gil Hong, Min Beom Lee, and Kang Ryoung Park, "Convolutional Neural Network-Based Finger-Vein Recognition Using NIR Image Sensors," Sensors, Vol. 17, Issue 6(1297), pp. 1-21, June 2017

 

===========================================================================================================================================================================================================

 

< DFingerVein-DB1 & CNN Model Request Form >

 

Please complete the following form to request access to the DFingerVein-DB1 & CNN Model (All contents must be completed). This database & CNN Model should not be used for commercial use.

 

Name :

 

Contact : (Email)

(Telephone)

 

Organization Name :

 

Organization Address :

 

Purpose :

 

 

Date :

 

               Name (signature)

 

===========================================================================================================================================================================================================

 

 

 

4. Dongguk Night-time Human Detection Database (DNHD-DB1) & CNN model

 

(1) Introduction

This database was obtained through a nighttime visible light camera (Logitech C600 web camera (https://support.logitech.com/en_us/product/5869)). DNHD-DB1 was constructed using images obtained from cameras fixed at various locations.

 

Table 1. Description of DNHD-DB1

 

DNHD-DB1

Number of images

Human

19760

Background

19760

#of channel

Color

(3 channels)

Width of human (background) image

(min. ~ max.) (pixels)

15 ~ 219

Height of human (background) image

(min. ~ max.) (pixels)

45 ~ 313

Environment of database collection

- Image acquisition using a static camera in surveillance environment

- The height of the camera from the ground was about 6–10 m

- Database was collected at various places (at 8–10 pm)

 

In addition, we made human detection algorithm in nighttime image by convolutional neural network (CNN) trained with DNHD-DB1. We also made this trained CNN model open to other researchers.

 

 

(2) DNHD-DB1 & CNN model Request

To gain access to the database and CNN model, download the following DNHD-DB1 & CNN model request form. Please scan the request form and email to Mr. Jong Hyun Kim (zzingae@naver.com).

Any work that uses or incorporates the database must acknowledge the authors by including the following reference.

 

Jong Hyun Kim, Hyung Gil Hong, and Kang Ryoung Park, "Convolutional Neural Network-Based Human Detection in Nighttime Images Using Visible Light Camera Sensors," Sensors, Vol. 17, Issue 5(1065), pp. 1-26, May 2017

 

===========================================================================================================================================================================================================

 

< DNHD-DB1 & CNN model Request Form >

 

Please complete the following form to request access to the DNHD-DB1 & CNN model (All contents must be completed). This database and CNN model should not be used for commercial use.

 

Name :

 

Contact : (Email)

(Telephone)

 

Organization Name :

 

Organization Address :

 

Purpose :

 

 

Date :

 

               Name (signature)

 

===========================================================================================================================================================================================================

 

 

 

3. Dongguk Body-based Person Recognition Database (DBPerson-Recog-DB1)

 

(1) Introduction

This database was collected from 412 persons with different body-view of subjects such as front, back, and side views. The cameras (visible light and thermal cameras) were placed near to each other to create a dual-camera set up, and placed at a height of roughly 6 meters. As the visible light camera, a webcam camera (C600) (https://support.logitech.com/en_us/product/5869) was used, and a Tau2 camera (http://www.flir.com/cores/display/?id=54717) was used for capturing thermal images. The visible camera captures images with a size of 800 × 600 pixels. The thermal camera captures images with a size of 640 × 480 pixels.

For each person, 10 images were respectively captured by visible light and thermal cameras to simulate the variation of body gaits. In total, we collected a database including 8,240 images (4,120 visible light images and 4,120 corresponding thermal images). Among the 412 subjects, there were 254 males and 158 females.

In addition, the following augmented images are included in this database.

 

Table 1. Description of the training and testing databases in our experiments

Database

Males

Females

Total

Training Database

Number of Persons

204 (persons)

127 (persons)

331 (persons)

Number of Original Images

4080 images (204×20)

2540 images (127×20)

6620 (images)

Number of Artificial Images

20400 images

(204×20×5)

10160 images

(127×20×5)

33100 images

Testing Database

Number of Persons

50 (persons)

31 (persons)

81 (persons)

Number of Original Images

1000 images (50×20)

620 images (31×20)

1620 images

Number of Artificial Images

5000 images

(50×20×5)

3100 images

(31×20×5)

8100 images

 

 

(2) DBPerson-Recog-DB1 Request

To gain access to the database, download the following DBPerson-Recog-DB1 request form. Please scan the request form and email to Dr. Dat Tien Nguyen (nguyentiendat@dongguk.edu).

Any work that uses or incorporates the database must acknowledge the authors by including the following reference.

 

Dat Tien Nguyen, Hyung Gil Hong, Ki Wan Kim, and Kang Ryoung Park, "Person Recognition System Based on a Combination of Body Images from Visible Light and Thermal Cameras," Sensors, Vol. 17, Issue 3(605), pp. 1-29, March 2017

 

===========================================================================================================================================================================================================

 

< DBPerson-Recog-DB1 Request Form >

 

Please complete the following form to request access to the DBPerson-Recog-DB1 (All contents must be completed). This database should not be used for commercial use.

 

Name :

 

Contact : (Email)

(Telephone)

 

Organization Name :

 

Organization Address :

 

Purpose :

 

 

Date :

 

               Name (signature)

 

===========================================================================================================================================================================================================

 

 

 

2. Dongguk Body-based Gender Database (DBGender-DB2)

 

(1) Introduction

This database was collected from 412 persons with different body-view of subjects such as front, back, and side views. The cameras (visible light and thermal cameras) were placed near to each other to create a dual-camera set up, and placed at a height of roughly 6 meters. As the visible light camera, a webcam camera (C600) (https://support.logitech.com/en_us/product/5869) was used, and a Tau2 camera (http://www.flir.com/cores/display/?id=54717) was used for capturing thermal images. The visible camera captures images with a size of 800 × 600 pixels. The thermal camera captures images with a size of 640 × 480 pixels.

For each person, 10 images were respectively captured by visible light and thermal cameras to simulate the variation of body gaits. In total, we collected a database including 8,240 images (4,120 visible light images and 4,120 corresponding thermal images). Among the 412 subjects, there were 254 males and 158 females.

In addition, the following augmented images are included in this database.

 

Table 1. Description of the training and testing sub-databases and the corresponding augmented databases for our experiments

Database

Males

Females

Total

Augmented database

Learning database

Number of persons

204 (persons)

127 (persons)

331 (persons)

Number of images

73,440
(204
× 20