The Research Of Image Retrieval Based On Color And Shape Features

In recent years,with the rapid development of information technology,the image information has been used more and more widely.The Content-Based Image Retrieval (CBIR) technique has already become a research focus.This paper introduces the background,the importance and the current research situation of CBIR technique briefly and discusses on retrieval of image color feature and shape feature deeply.The main work has been given as follows:On the aspect of color feature exaction,this paper proposes a retrieval algorithm based on sub-block weighted color histogram through analyzing the different color space,the quantization of color and the common method of color feature exaction,which considers the visual features as well as the distribution features of spaces.Firstly the image is converted into HSV space, then it is divided into several sub-blocks using the block algorithm and uses the visual weight of the complexity of color in weighted disposing each pixel point in every sub-blocks.This paper also constructs a weighted color histogram to be the color features of this sub-blocks and matches in summary in order to achieve a better result.On the aspect of shape features extraction,this paper gives the typical feature extraction algorithm and analysis their limitations too,proposes a shape feature extraction based on Contourlet transform,integrates the features of color and shape,and proposes an image retrieval algorithm based on neural network in integrating features weights. Considering the difference of physical meaning and value ranges, this paper adjusts the features weights at the same time of normalization the features of color and shape weights,converts the adjustment problem of Integrating the features of image retrieval into optimization problem,and uses RBF neural network in learning for getting the optimization values to realize retrieval accurately.This paper is not only giving the related algorithms, but also carrying out related experiments,verifies the validity and superiority of algorithms.