Study Of Image Retrieval Based On Fusion Of Color And Texture Features

With the rapid development of the computer, communication technique, multimedia and network technology, there are more and more resources of digital images available. How to search the image data in the massive database fast and effectively has attracted more and more people. Now the Content-Based Image Retrieval technique has been a hot research subject.This thesis firstly introduces the related research on CBIR both at home and abroad and the existing problems. Then this paper introduce the key technique in CBIR. During the study of color features extraction and matching, HSV color model is chosen and divided into small spaces according to the perception of human eyes. The details of extracting HSV-based color histogram are described. However, the color histogram includes no spatial distribution information of image and the conventional distribution pays no attention to the main body of the images or the interested areas, nor does it consider the relations among the different parts. In order to deal with all the shortcomings, this pager has made some modifications on the algorithm of color feature. By improving the color distributing approach, this paper proposes an image retrieval method based on color-space histogram, which designs rectangular overlapped sub-regions to get color histogram in each region. Then the feasibility for the above extracting color feature algorithms is also validated by the experiment. In order to use multiple features to retrieve images more effectively, this paper has studied the image retrieval method using color feature and texture feature. Co-occurrence matrixes are good at extracting important information from the whole images and Gabor filters are good at extracting information in frequency and direction from local area of images. This paper puts forward combining Co-occurrence matrixes, Gabor filters and the improving color distributing approach to extract images features. It is validated by the experiment that better image retrieval performance can be achieved by combining the three features. Finally, the paper summarizes the research on CBIR and proposes the future research orientation in this area.