Person re-identification using prioritized chromatic texture (PCT) with deep learning

Abstract

Person re-identification (re-ID) helps to identify a person’s attention in different cameras. But this is not an easy task, due to distance, illumination and lack of dataset. Nowadays, this field attracts many researchers because of its varied applications. Here, the information of both local texture and global color representations are concatenated with an original raw image. This concatenated information is gathered by finding the maximum value of chrominance in terms of HSV, texture in terms of Scale Invariant Local Ternary Pattern (SILTP) for each pixel and original raw image. SILTP is well known for its illumination invariant texture description. Convolutional Neural Network (CNN) is used in the proposed work to extract the features from the concatenated information. The proposed Prioritized Chromatic Texture Image (PCTimg) is concatenated with original raw image and fed into CNN. Here, finally a six dimensionalfeature is fed into CNN to extract the deep features. Cross-view Quadratic Discriminant Analysis (XQDA) similarity metric algorithm is employed to re-identify a person.Multiscale Retinex algorithm is used for pre-processing the images.To address the challenges in terms of view point deflection, a sliding window is formed for describing local details of a person in the SILTP feature extraction phase. The HSV helps to incorporate the human color perception. The triplet loss function is used to learn the similarity and the dissimilarity of the training images. The performance analysis of the proposed work is improvedwhen compared to the existingworks.