Evolutionary optimization of JPEG quantization tables for compressing iris polar images in iris recognition systems

Recognition performance in iris biometrics strongly depends on the image quality. The appliance of compression algorithms to iris images raises the question whether it is possible to adapt those algorithms for biometrical purposes. In this work, we propose customized JPEG quantization matrices for compressing iris polar images to positively impact the recognition performance. We build on previous research and apply a genetic algorithm to obtain specialized matrices for destined compression ratios. The proposed tables are able to clearly outperform JPEG's standard quantization matrix. Moreover, some matrices also provide superior results in terms of ROC characteristics as compared to the reference scenario using uncompressed images. This leads to clearly lower error rates while also significantly reducing the necessary amount of data storage and transmission.