Research On Multi-Features Based Image Retrieval Technology

Content Based image retrieval (CBIR) is a technique which retrieves images by the content features of the image. Since the 90s of last century to now, it has become a research focus at home and abroad. This paper describes the background and current situation of the content-based image retrieval first; and CBIR are discussed in more detail the basic principles, methods and feature extraction, feature description, feature matching, similarity measure and other key technologies.As Color, texture and shape is the three features which is most commonly used in CBIR, This paper deeply studied the image retrieval technology based on these three single features, then it mainly studies the image retrieval technology integrated these three features, It implemented a image retrieval method of comprehensive more features.Color space and its quantization, the color feature extraction and similarity measure are the key technologies based on the color feature in image retrieval. In the Quantization of color space, the use of non-equal interval to quantify the value of feature extraction methods in order to better adapt to the characteristics of the human eye color perception. It uses the histogram, cumulative histogram, color aggregation vector and color moment to describe and extract color features. In the extraction of texture feature, Due to the gray level-gradient co-occurrence matrix combined the gray information and gradient information of the image feature, it extracted the contour feature of the image first, and then it descripted the texture feature by the contour feature and gray feature for texture features that contain more comprehensive information. In the extraction of shape feature, it extracted the image moment invariants to segmentate images by using the Zernike moment invariant method, which is based on the unequal spacing ring block. Its edge detection is based on using the Zernike operator, it calculate gradient using one-order differential coefficient of the gaussian function by locating the local maximum of image gradient. It analyzes and detects strong edges and weak edges by two thresholds of algorithm. In the choice of multi-feature combination way, it based on an overall consideration of the principle of combining space information, statistical information and shape information of the image. It contains the most comprehensive image information to the the feature vector of the multi-features using the linear method of adding. It matches the multi-features using a fuzzy feature matching algorithm for similarity, thus increasing the recall rate of retrieval.The relevance feedback technology is introduced into the retrieval system, it reduced the gap between the low-level features and the high-level semantic concept, increased the accuracy of retrieval system by modifying the query vector feedback method, modify the weights of similarity measure.