Research On Similar Image Retrieval Based On Deep Learning Algorithm

For the past few years, with the rapid development of multimedia and Internet technology, the spread and storage of digital products is becoming more and more convenient, especially for digital video and digital image. According to statistics, an average of more than 20 hours of video are uploaded to YouTube and about 100000 images are uploaded to Facebook per minute. Such a large amount of data for the traditional similar image retrieval algorithm is a great challenge. The most used algorithms of similar image retrieval are based on the content of similar image search (CBIR), but this kind of algorithm just can extract the low level features of images(such as color, texture, shape), and it is very difficult for them to extract the high-level image features. In order to obtain the high-level semantic feature of images and improve the ability of similar image retrieval, the main research works of this paper can be summary as following:1) Through exploring the theory of deep learning, This paper analyzes the effectiveness of applying the deep learning algorithm in image feature extraction, and analyzes the feasibility of applying the deep learning algorithm in similar image retrieval.2) This paper designs a similar image retrieval model based on deep learning, and elaborates the details of modeling. At last, through analyzing the weight images of the deep learning model, this paper verifies that the deep learning has sparsity. Therefore, it can obtain the high-level structural feature of images and improve the ability of image feature representation.3) This paper proposes a new method for similar image retrieval based on deep learning algorithm, and verified by the experiments in binary images retrieval, our method is superior to the traditional content-based image search algorithm and the perception hash algorithm in both the precision and recall.