Research On Content-based Image Retrieval And Relevance Feedback Technology

Along with the rapid development of computer multimedia technology and Internet,image retrieval is widely applied to the communication technology, space exploration,industrial production, and medical and other fields. Now the image information are increasingquickly with the advent of the era of big data. How to find the query image which meet thedemand of people in the vast amounts of image information more quickly and efficiently,promote the development of image retrieval technology.In order to obtain image information better, the technology of content-based imageretrieval has become one of top research topic, it is to use the color, shape, texture of imageand other low-level visual features for image retrieval. In this paper, we use two kinds offeatures, one of is color histogram which is commonly used, the second one is a novel methodtexture histogram; it is a combination of color and edge direction at the same time to describethe information of the image. Since the HSV color space more conform to humans’ visualperception mechanism, this paper will first converted the query image from RGB color spaceto the HSV color space. On the other hand, in order to reduce the "semantic gap" between thelow-level features and high-level semantic, to join the relevance feedback mechanism offeature re-weighting. In the phase of related feedback, considering the related sample andno-relevant sample, in order to maximize the best benefits relevant images from thenon-relevant ones, we introduced a new kind of feature re-weight method to achieve a bettereffect of image retrieval.Finally, we experiment on two data sets which is often used in image retrievalexperiments. The experiment results show that the two kinds of different feature in relevancefeedback method based on the feature re-weighting, have achieved very good effect. The newfeature re-weighting method we introduced is significantly higher than the other two kinds ofweights on the precision and recall of retrieval, the paper proves the effectiveness of the newfeature re-weighting method.