Research On Face Image Retrieval By Combination Global Feature And Local Feature

Content-based image retrieval is a hot topic in the field of image analysis. As the main basis for authentication, image retrieval technology on Facial image which can be applied in various fields, has important significance.Face general characteristics can be described by local or global features, global features can be integrated considerations all parts of image, and local features can reflect the details of local information. However, the human face is so complex that, it is difficult to retrieve facial images efficiently and accurately by a single type of feature.In this paper, we proposed a new content-based retrieval algorithm for facial image retrieval. In the algorithm, face images are retrievaled by a combination global features and local features. The main work is as follows:Firstly, in the module of faces retrieval by local features, we use Dense Sift as local feature descriptors, and use BOW to calculate the characteristics of the input image and sample image set. After that we could get a similarity sequence based on local feature. Secondly, in the module of faces retrieval by global feature, we use deep neural network to extract the abstract characteristics of facial images, and construct an index by hierarchical clustering tree which can help us get another similarity sequence based on global feature. Finally, we combine the two similarity sequences by setting weights to obtain a final similarity sequence, and output the search result from the sample database according to the similarity sequence.Experiments in this paper show that, compared to the traditional algorithms, the algorithm we proposed more accurately retrieve the image we need.