Person re-identification method based on CNN and manually-selected feature fusion
Haohua Ku, Ping Zhou, Xiaodong Cai, Haiyan Yang, Yun Chen · 2017
This paper presents a method based on Convolutional Neural Networks(CNN) and manually-selected feature fusion to improve the accuracy for person re-identification. Firstly, features of input images are extracted based on Inception structure. Secondly, a manually-selected feature module is added to adjust the learning of the Inception network for obtaining fusion features which are used for a softmax classifier. Experimental results indicate that compared with the networks without Inception structure, the proposed network on the VIPeR dataset improves the Rank-1 accuracy by 3.97% on LFDA.