Study On Image Retrieval Method Based On Improved Color Difference Histogram

With the advancement of computer and Internet technology, the number of digital images rapidly increase. How to quickly and efficiently find the image which meet the need of people in the tremendous amounts of image database, and accelerate the research of image search technology.In order to improve retrieval performance, content-based image retrieval（CBIR） is becoming more and more popular.So far,content-based image retrieval use low-level visual features in image retrieval. In his article,we use texture,shape, color in image retrieval. Firstly, color histogram and edge-orientation are extracted from image in CIE L*a*b* color space. this paper unites color and edge-orientation to acquire the color difference histogram（CDH） feature Secondly, in order to use the information of image fully, and improve retrieval performance,color co-occurrence matrix feature is extracted from image. Finally, this paper unites color, edge-orientation and color co-occurrence matrix to obtain the joint histogram feature（JHF） for image representation.In this thesis, we experiment on Olivia₂688, Corel₅000 and Corel₁0000, which are prevalent in image retrieval experiments. The results show that the proposed descriptor in this article is much more efficient and effective in these database. At the same time, to illustrate our experiment is universal, We conducted a comparative experiment with the existing image feature descriptors, for example, color difference histogram, micro-structure descriptor. The experiment results demonstrate that it is much more efficient and common than representative feature descriptors in precision and recall rate. The feature descriptor of proposed is effectiveness and universality. In addition, the image retrieval link distance metric, Euclidean distance, Manhattan distance metric and improved Canberra distance also carried out comparative experiments, the experimental results show that the improved Canberra distance effectively improve the retrieval precision and recall.Finally, by using the software engineering knowledge we achieved the image retrieval system. the user can choice the database of the image retrieval freely, and the user is easy to use it.