Retrieve Images Based On Color And Shape Features

With the development of multimedia resource such as image, the traditional retrieval methods can no longer satisfy the requirement on retrieval. Content-based image retrieval technology emerges in this circumstance, which is convenient to retrieve image. With the analysis on content-based retrieval technology, this paper would exhaustively analyze the existing feature extraction methods from color feature, texture feature and shape feature respectively, constructing practical image retrieval system. Then this paper would give further analysis on a novel image retrieval algorithms based on color and shape. The research content is as follows:First, this paper would analyze and study image extraction from aspects of color feature, texture feature, and shape feature; and give some further explanation on common extraction methods.Second, this paper would demonstrate a novel color-based image retrieval algorithm that combines prominent interest points, color moments, and distance histogram. This is a method that takes user-concerned interest point as the main visual cue and combines the color feature to retrieve image through three steps:the detection of prominent interest points, feature extraction of distance histogram, and feature extraction of color moments. Taking both local color feature and spatial distance relation of interest points into consideration, this method overcomes the defects of traditional color moments lacking of location information. The experiment results show that this method is easy to use and successfully improves the efficiency of image retrieval.Third, this paper presents a novel shape-based image retrieval algorithm. On the basis of image contour extracted by using GAC model of partial differential equation, the Hu invariant moments could be computed, Which would be the image feature descriptor for image retrieval. As to numerical implementation, there would be discussion on Gaussian function and differentiate operation; according to which, the smooth processing of image and the computation of image gradient would be simplified, and the computing speed would be improved. The experiment results demonstrate that this method is scale and rotation invariant.