Research On Person Re-identification Based On Image Saliency And Bag Of Words Model

Person re- identification is the process of re- identify the specific person target from all the person targets that recognized from pictures. In recent years, it has been a popular direction in computer vision and pattern recognition. It has wide application prospect in the fields of intelligent video surveillance, content-based image retrieval, intelligent transportation and advanced human-computer interaction.In the use of the bag of words model for person re- identification, we introduce image saliency detection for each picture to generate their own weight matrix to highlight the human body. The use of the Gauss template can only be suppressed the impact of the background to a certain extent, because the human form is irregular and not all in the center area. And the image saliency detection can highlight the interested region of the image, that is the human target in this experiment, and thus better inhibit the impact of background.So, the person re-identification based on image saliency and bag-of-words model is proposed. First of all, the image is detected and a significant weight matrix is generated. Then the features of each image are extracted. The features of pixels in the salient regions are assigned a higher weight value, and lower weight value are assigned to the background region. In additio n, we introduce multiple queries to adapt to the extensive image variations and improve the matching accuracy.Finally, we experiment on the Market-1501 test set. The results show that our method makes the average matching precision increased from 19.20% to 20.14%, and the rank-1 increased from 42.14% to 43.77%.And proved that the method in this paper is feasible.