Bayesian sample steered discriminative regression for biometric image classification

摘要: The flow chart of the proposed BSDR algorithm. Using method described in Section 3.2, target response vector for each class k can be computed. Then linear mapping function obtained. With all w ? computed, matrix obtained by W= 1 , ?, K . For a test sample y, its labeling feature extracted ry=WTy. Classification done NNC with cosine distance. A novel image extraction is proposed.BSDR refers to Bayesian steered discriminative regression.BSDR both uses label and appearance learning.Experimental results demonstrate effectiveness method. Regression techniques, such as ridge regression (RR) logistic (LR), have been widely used supervised learning pattern classification. However, these methods mainly exploit information learning. They will become less effective when number training samples per small. In visual classification tasks face recognition, images also conveys important information. This paper proposes based model, namely (BSDR), which simultaneously exploits virtue formula. learns extract features, simply nearest neighbor classifier. has advantages small mappings, insensitiveness input dimensionality robustness size. Extensive experiments on several biometric databases promising performance our