Evaluation of image compression algorithms for fingerprint and face recognition systems

A variety of widely accepted and efficient compression methods do exist for still images. To name a few, there are standardised schemes like JPEG and JPEG2000 which are well suited for photorealistic true colour and grey scale images and usually operated in lossy mode to achieve high compression ratios. These schemes are well suited for images that are processed within face recognition systems. In the case of forensic biometric systems, compression of fingerprint images has already been applied in automatic fingerprint identification systems (AFIS) applications, where the size of the digital fingerprint archives would be tremendous for uncompressed images. In these large scale applications wavelet scalar quantization has a long tradition as an effective encoding scheme. This paper gives an overview of the study BioCompress that has been conducted at Fraunhofer IGD on behalf of the Federal Office for Information Security (BSI). Based on fingerprint and face image databases and different biometric algorithms we evaluated the impact of lossy compression algorithms on the recognition performance of biometric recognition systems.