Research On Algorithm Of Content-Based Image Retrieval

With the rapid development of the multimedia techniques and Internet, the amount of multimedia information is increased rapidly. Therefore, the rapid and effective management for large-scale image databases becomes an important research topic. In this research area, image database system plays an important role in the multimedia information system because of its abroad employment in many important applications. So, currently, rapid and effective searching for desired images from large-scale image databases becomes an important and challenging research topic. Content-based image retrieval (CBIR) is the set of techniques to address the problem of retrieving relevant images from an image database based on automatically derived image features.Based on understanding of the content on CBIR, and researching its related field's technology, this paper sets focus on the feature extraction of image and technology of relevance feedback. The main research ceontent of this paper include the following aspects:Firstly, the HSV color histogram can't describe spatial information effectively. The traditional description to local color features neglected the contacts of every parts and neglected the difference between images when people observing. So this paper improved retrieving algorithm of color feature based on the dominant colors of mufti-resolution partitions. By the experiment, it can solve above question well, and adjust the size of central regional according to user need.Secondly, in order to use Multiple Features in image retrieval system, this paper research on extracting texture features of images using Gabor wavelet, and improved extraction algorithm of Texture feature. To represent the color content of an image, the color indexing approach using the dominant colors of multi-resolution partitions is used. To represent texture feature, Gabor wavelet is computed. Through the image retrieval experiment, better image retrieval performance can be achieved by combing two kinds of features.Thirdly, at present there are some key problems on content-based image retrieval techniques focusing on the gap between low-level features and high-level semantics. A relevance feedback scheme based on feature weight adjustment and Bayesian theory for image retrieval is proposed. It realized a color feature re-weighting approach and ranks the image in term of the predictive probability in the next feedback. With plentiful experiments, it is proved that the efficiency can be enhanced by using this method.