Multispectral Morphing Dataset

Description:

Multispectrl Morphing Dataset (MSMD) comprised of 143 subjects (86 males and 57 females) with varying age groups from 22 to 72 years. The MSMD dataset was collected in multiple sessions to capture visible and multispectral images from 143 subjects.

Visible Data collection: 

The visible images are collected using DSLR camera (Model: Nikon D320) under constrained conditions. Because visible images represent passport images, special attention is required to achieve high-quality face image capture. Visible images were captured using a photo booth with uniform lighting, neutral pose, and expression. Visible data were captured in two sessions with a time gap of 30- 45 days. Sesison-1 was collected under highly constrained conditions, while session-2 was collected under similar conditions. Session-1 is used as the enrolment samples and session-2 samples are used as the trusted capture in the experiments. In session-1, 13 images were captured and in session-2, 30 image samples were captured. Thus, the visible dataset comprised 143 subjects x 13 samples = 1859 samples in session-1 and 143 subjects x 30 samples = 4290 samples in session-2. 

Multispectral data collection:

The multispectral data were collected using a custom device built using a CMOS camera (Model: BCi5-U-M-40-LP) with 1.3 Mega pixels spatial resolution. In this study, we selected six different spectral bands, 650 nm, 710 nm, 770 nm, 830 nm, 890 nm, and 950 nm, together with the WholeLight (WL) image. We selected these six spectral bands to cover both visible (VIS) and near-infrared (NIR) wavelengths. Furthermore, the availability of the WholeLight image, in which the image is captured without any filters, represents the standard image captured with the VISNIR-sensitive camera. The dataset is collected in the indoor setting with two illumination units of a quartz tungsten halogen (QTH) light source. We have collected 11 samples for each data subject that will result in 143 subjects x 7 spectral bands x 11 = 11011 samples. 

Morphing Images:

The morphing images were generated using the visible images collected in session-1 of the visible image dataset. We selected one image per data subject and employed two morphing generation methods to create morphing attack samples. The first morphing generation is based on using landmarks in which pixel-level information from the contributing data subjects is employed by wrapping and blending to generate high-quality morph images. The second morphing generation is based on a generative method called MIPGAN-2, which uses GAN-2 as the backbone. The motivations behind the selection of these methods include (a) high-quality morphing image generation, (b) indicating high vulnerability with FRS and (c) challenges to be detected using MAD techniques. To ensure a disjoint-morphing dataset, the dataset of 143 unique subjects was divided into two disjoint sets. The training set consisted of 78 data subjects and the testing set consisted of 65 datasets. Morphing is performed internally on the training and testing sets to achieve a complete disjoint set. In total, we had 1400 morphing images, of which the training set consisted of 928 images and the testing set consisted of 472 morphing images.

Following figure shows an example of morphing the images from the MSMD dataset

The statistics of MSMD dataset summarising the data partition for training and testing set is  as follows:

Copyright of MSMD :

 Researchers can avail Multispectral Morphing Dataset (MSMD) by following the procedure mentioned: 

• Researchers are required to send the request to use MSMD database by sending acopy of the license agreement, completely filled, to vetrekarnarayan@unigoa.ac.in with the subject line "License agreement for Multispectral Morphing Dataset". 

Note: The license agreement has to be signed by the researcher or supervisor and the signature of a legal authority on behalf of the institution, such as the Head of the institution or Registrar, along with the institutional seal (Rubber Stamp).  The license agreement should be on the researchers institutional letterhead. 

The request will not be considered if due procedures are not followed.  

• Every request to avail 3D-PCPA database will be placed before the Ethical Committee of the Institute for approval and the confirmation of the request will be sent via email. Further, all the instructions related to access to the database will be detailed in the email.

This database is available only for research and educational purposes and not for any commercial use. Further, the use of the database is only for one year from the date of signing the agreement by the researcher. Beyond the allocated one year for the usage of data, the applicant has to apply again following the due procedure.  

The database is available only for research and educational purposes. All the rights of the Multispectral Morphing Dataset (MSMD) are reserved, and commercial use/distribution of this database is strictly prohibited. All the technical reports and papers that report experimental results from this database should provide due acknowledgement and reference. If you use the database in any publications or reports, you must refer to the following paper: 

• Raghavendra Ramachandra, Sushma Venkatesh, Naser Damer, Narayan Vetrekar, R. S. Gad, Multispectral Imaging for Differential Face Morphing Attack Detection: A Preliminary Study, IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pp.6185-6193, 2024