The Research On The Application Of Color Feature In Image Retrieval

In the Content-based Image Retrieval, color as one of the most important visual information has been widely used. It reflects the gray scale information of component of the color space in pictures taking the form of histogram, and it has such features as simple, perspicuity, and interference killing etc. But as the histogram is prone to be influenced by illumination, the problems of content constancy have appeared. For this, we put forward a new algorithm of qualitative matching of histograms based on scale-space theory. We have deeply studied the theory of scale-space, and put it into use of histograms of images. First, creates the scale-space of histograms, and extracts qualitative features of each histogram, then gets the matching measurement based on prescriptive formula. On account of this algorithm taking time in image pre-handling, so we advance an algorithm of layering matching of histograms based on scale-space theory. It matches histograms of the scale-space from the coarse scale layer to the fine scale layer, until retrieves the most matching images. As to testify the validity of the algorithm, we create a CBIR System based on image histograms of scale-space, and use it to carry out some experiments on the algorithm of qualitative matching of histograms and the algorithm of layering matching of histograms based on scale-space theory. At the same time, we have put these algorithms in Gray,RGB and HSV Color Space, also compared them. In the experiment, the algorithm's recall ratio, precision ratio, the time of pre-handling and query time were calculated respectively, and compared with the conventional method of color histogram intersecting. The results bear out that our algorithms have the excellent performance, such as not sensitive to the picture's transition, rotation and size change and have little to do with the size of database, also more excellent than the conventional method of color histogram intersecting.