Research And Application Of Large Scale Face Image Retrieval

With the rapid development of economic society and Internet technology, digital media technology changes quickly. There exits continuous innovations in the fields of computer vision and pattern recognition. Intelligent information processing technology has been widely applied. Accompanied with it are the tens of thousands of image data resources. Facing with the large size of the image resource, how to accurately and quickly retrieve the image information from the large image database becomes a very practical subject which is full of theoretical value. Large-scale face image retrieval as one specific application has important research value and practical significance. Due to the limitation of experimental conditions, this topic is carried out in the massive experimental environment with 200 thousands face images.Given a face image as a query, the goal of Large-scale face image retrieval system is to retrieve images containing faces of the same person appearing in the query image from a web-scale image database containing tens of millions face images. The retrieval performance has a great relationship with the feature extraction of face image and the quality of coding. The Performance of traditional human face image retrieval system is not very ideal in character coding, retrieval speed, retrieval precision and so on.For the above problems and the laboratory project requirements, in this paper, we have learned the relevant knowledge of large-scale face image retrieval. And finally we complete the system design and implement the large-scale human face image retrieval system. Compared with the previous systems, this system is improved in the following ways: First, Image coding will take up less memory space. Besides, with multiple references reshuffle operation, the effect of Image retrieval system is better. The main contents of this thesis are as follows.1. After collecting the facial image and preprocessing the face image, local feature extraction operation will be done in order to facilitate the similarity matching of facial image and the effective use of memory space. In this thesis we uses a coding method which is based iteration algorithm VLAD of extracted features vector encoding polymerization, making the resulting polymeric descriptors have a stronger dimension reduction capability and better query performance.2. When researching the face image retrieval, we find local features are complementary to global features. In the retrieval process, we use the algorithm of asymmetric distance calculate inverted index to make index on the local features which is aggregated by VLAD. By traversing the inverted index table retrieves the candidate face images. The retrieve result has a lower accuracy and higher recall rate. While global features combining the technology of multiple references shuffle retrieve the candidates face images again, to improve the human face image retrieval results further.