Finger Vein Sample Compression using Recent AI-Based Still Image Compression Schemes

In the paper "Finger Vein Sample Compression using Recent AI-Based Still Image Compression Schemes" we present an evaluation of recent AI-based image compression schemes applied to finger vein samples from 3 different datasets. Due to the page limit of the original paper, not all plots are contained in the paper. Here you can find all the plots as well as the score files of the evaluations for all three evaluated datasets and all the different feature extraction and compression schemes.

Abstract

The compression of the sample files enables more efficient storage and transmission of biometric data. Four different recent still image compression techniques (JPEG2000, JPEG XL, JPEGAI and CompressAI) are applied to finger vein sample images in two different scenarios, first where the gallery is uncompressed and second where the gallery is JPEG2000 compressed (as it is the current ISO/IEC standard). The impact on the recognition accuracy is assessed on three different data sets using six finger vein recognition schemes to answer the question of which compressing works best for the probe samples in each of the two scenarios. The results show that JPEG2000 is not competitive, but the more recent methods, especially JPEGAI, perform well up to a compression ratio of 200 (0.04 bpp).

Plots

The following file contains all the EER/FMR1000/ZeroFMR as well as the Image Quality Assessment Evaluation plots as eps files (generate by MATLAB)
Each single plot contains both scenarios, scenario 1 with the uncompressed gallery images and scenario 2 with the J2K compress gallery images.

Score Files

The following file contains all the evaluation scores (EER/FMR1000 and ZeroFMR) for the different image compression schemes.
Each file is a MATLAB .mat file containing a MATLAB table with the values per feature type, dataset and image compression scheme.
The scores for the image quality assessment methods are included as well (also as a MATLAB table).