Person Re-identification Based On Saliency

It requires a large amount of human efforts on exhaustively searching for a person from a lot of images and vedios.Researchers call this problem that matches pedestrians observed from disjoint views person re-identification.A person observed in diffirent camera views undergoes siginificant viraitions on viewpoints,illumination,and poses,which make great variations in image apperance.This paper presents a novel perspective for person re-identification based on learning human saliency.We presents a feature representation method based on superpixels in the stage of feature extraction.Firstly,the pedestrian image is divided into several superpixels.Then the feature space is constructed by combining the color histogram and the speed up robust feature.Experiments show that the combination of the feature descriptor and the existing matching method can greatly improve the efficiency and effectiveness of person re-identification.The saliency of patches is weighted by the difference among its matched ones in traditional person re-identification methods.However,the result is difficult to keep steady with the change of samples.In this paper,we introduce cellular automata to intuitively detect the inherent salient object in pedestrian images.In order to take advantage of the superiority of above-mentioned two methods,We propose an effective method to incorporate multiple saliency maps with multi-layer cellular automata.Experiment shows that it has improved the performance of person re-identification.Finally,it is crucial to determine the importance of each feature in matching process.The traditional feature selection is too simple which selectively assign weights to informative feature according to its differences among appearance attributes.Therefore,a learning sorting method is proposed to measure the similarity between the images.The experimental results show that the proposed approach outperforms the state-of-the-art person re-identification methodson the i LIDS.