Research Of Image Retrieval Based On The Fusion Of Local And Global Features

With the rapid development of multimedia technology and communication technology, digital image acquisition and storage is more and more convenient, their quantity showed a trend of explosive growth. However, the large amount of image were in a state of disordered distribution, the information contained cannot be used efficiently. Therefore, how to quickly find the interested image in huge amounts of images has become the urgent needs of people.The main research work of this paper is put forward a serial and parallel combined image retrieval system, of the system comprehensive utilize the global features、local features of the image and image perception hashing technology, the recall ratio precision ratio in retrieval and retrieval efficiency can be achieved better effect, the specific content as follows:1. Study the global and local features of images. Aimed at the local features, choose SIFT features and puts forward an improved SIFT algorithm, changed the block way of the descriptor area. That will reduce the feature vector by 128 d to 32 d, reduce the complexity of the algorithm. Then study the color global features, choose color histogram, and puts forward the concept of color information entropy, used for the features fusion of the image.2. Propose a feature fusion algorithm based on color information entropy. When fuse feature and measure similarity, dynamically allocated weights of global features and local features according to the color information entropy. That can better adapt to changes in the types of image and image library update.3. Propose a serial and parallel combined image retrieval system, the system uses a classification retrieval model. Primary retrieval use image perception hashing technology retrieval, can quickly reduce the retrieve range, improve the efficiency of retrieval. Secondary retrieval use the method of global features and local features, can improve the accuracy of retrieval.4. Establishment a serial and parallel combined image retrieval system and compared with other algorithms, three performance evaluation indicators the recall ratio, precision and the retrieval time verify the effectiveness of the system.