Logistic similarity metric learning for face verification

摘要: This paper presents a new method for similarity metric learning, called Logistic Similarity Metric Learning (LSML), where the cost is formulated as logistic loss function, which gives probability estimation of pair faces being similar. Especially, we propose to shift decision boundary gaining significant performance improvement. We test proposed on face verification problem using four single descriptors: LBP, OCLBP, SIFT and Gabor wavelets. Extensive experimental results LFW-a data set demonstrate that achieves competitive state-of-the-art verification.