A Fast Retrieval Algorithm for Person Re-identification Based on Hash

Abstract Aiming at the problem of slow retrieval speed due to a large amount of data in person re-identification, this paper introduces Taylor expansion into a deep neural network, and obtains Taylor deep neural network for feature extraction, which is combined with the proposed hash probability coding algorithm, and get a fast retrieval algorithm for person re-identification based on the hash. The algorithm can not only extract the feature map information well but also perform image retrieval at a faster speed, which is a new method to solve the fast image retrieval of person re-identification. First, a Taylor deep neural network framework is proposed, and the feature representation extracted by each layer of the deep network is fused by Taylor expansion weighting to retain the overall features of the image and make the feature representation more comprehensive ; then, the triplet loss function Triplet Loss is optimized to improve the When the distance between groups cannot be optimized, the group loss function Group Loss is proposed. Finally, a new hash probability compression coding method is used to make the image retrieval fast. Experiments show that the algorithm model in this paper is 26.7% and 8.18% higher than the BaseLine model on the large data sets Cuhk03 and Market1501, respectively. It also has a good retrieval speed on images retrieved by person re-identification..