基于图像描述的人物检索方法

Retrieve specific person in the surveillance scene is an important and urgent demand on the security field. In recent years, the method of image retrieval is mainly based on image content, which requires the image to be retrieved works as input and thus can’t meet the actual needs of surveillance and security. We propose a method of person retrieval based on image caption and present a new surveillance dataset SPCD which tags person caption. We verify the method in new dataset, the accuracy of gender, dress color and action respectively reaches 86.5%, 93.5% and 65.5%. This paper provides an effective way for application of retrieval person in the surveillance scene.