Criminal Investigation Image Retrieval Based on Deep Learning

This paper proposed a criminal investigation image retrieval algorithm based on deep learning. First, this paper improves Faster R-CNN object detection network making it can extract multi-scale features of images, so that it can better annotation semantic information for the image. To obtain a preliminary image list through image retrieval of the semantic information. Then use ResNet50 to extract image features, perform content-based retrieval in the image list returned by the first step of semantics to obtain the final criminal investigation image retrieval result, both the semantic and the content are combined for image retrieval. The experimental results show that the criminal investigation image retrieval algorithm proposed in this paper is superior to the simple content criminal investigation image retrieval algorithm in Precision and Retrieval rate.