An improved similarity metric based on joint Bayesian for face verification

The joint Bayesian method is an effective method for face verification. However, it is insufficient distinguishability for this method to calculate the similarity between different samples. To address this problem, in this paper, we proposed an improved similarity metric based on joint Bayesian for face verification. The proposed method defines two novel similarities for two samples based on joint Bayesian framework. Then, we fuse these two similarities using different weights to produce a decision similarity, and the weights are obtained by logistic regression. New similarity can achieve more distinguishability to different faces. The experimental results on challenging LFW and WDRef face datasets show the superiority of the proposed method for face verification tasks.