On comparison score fusion of deep autoencoders and relaxed collaborative representation for smartphone based accurate periocular verification
Raghavendra Ramachandra, Kiran B. Raja, Christoph Busch · International Conference on Information Fusion · 2016
Periocular biometric verification on the smartphone is gaining more popularity with the increasing functionality and accessibility of the services offered using smartphones. However accurate periocular verification requires especially on the smartphone to address several challenges due to the uncontrolled image capturing conditions. In this paper, we present a novel scheme for accurate periocular recognition based on a comparative score fusion. Given the periocular image, the proposed method will first perform the pre-processing to enhance the quality of the captured periocular image. The enhanced periocular image is processed further to extract texture features using Log-Gabor filters. In the next step, we train two different classifiers such as deep autoencoder and Relaxed Collaborative Representation (RCR) on the texture features. Then, given the probe periocular image, we obtain the corresponding comparator scores for each of these comparators that are fused using weighted sum rule to perform the verification. Extensive experiments are carried out on a large-scale smartphone based periocular database comprised of 550 subjects recorded in three different scenarios using three different smartphones. In depth analysis on the benchmark of the proposed scheme with five different state-of-the-art schemes demonstrates the outstanding performance of the proposed scheme with a consistent GMR of about 99% at FMR = 10−3.