Active learning for person re-identification
Jin-Peng Xiang · 2012
Person re-identification is defined as to find the same person who re-occurred in a multi-camera surveillance system. Existing machine learning approaches focus on extracting or learning discriminative features followed by template matching using a distance measure. However, labeling images for a training set is a time consuming task. In this paper, the person re-identification is considered as a binary classification problem. The active learning framework with SVM is applied to person re-identification problem in this paper. Rather than learning from all the training samples, the proposed method selects the most valuable sample according to the current knowledge of the classifier. Experimental results show that our proposed method not only can reduce the number of sample labeling but also achieve a higher accuracy with using less training samples