Research On Color-based Image Retrieval And Relevance Feedback System

With the popularization and development of Interner,the number of image data grow dramatic fast,and how to retrieve image efficiency and quickly become an important issue in the field of image's application.Text-base image retrieval(TBIR) eppeared in the 1970s,with the limit of artificial marking the feathers of images,content-based image retrieval(CBIR) become more absorted.CBIR extract low-level visual feathers as the retrieval feathers,such as color,texture,shape etc. Generally,an example image is needed when a user use the CBIR system,which can extract the feathers of the example image, then compare with the feathers of other images in the database and show the resut to the user.This paper first discussed how to extract the color as the feathers.There are three methods to extract the feathers: Dominant Hue Method(DHM), Global Color Histogram Method(GCHM) and Dominant Color of Partition Method(DCPM). DHM divide the color space into several areas, suit goal and background color relatively simple image, but this method unable to catch the spatial relationship between the main colors. GCHM is similar toDHM, but it pays close attention to each color appearing in the image, so the method is suitable for the color of the image comparely mixed and disorderly,This method does not pay close attention to the spatial relationship too. While DCPM pay close attention to space distribution of color . the basic thinking of thist method lies in that if two images have same color in the same space, they have greater similar degree.Later discussed the realization of relevant feedback of the above method . The goal of relevant feedback system lies in making the system have the ability of learning, so that ia can find and catch users' actual inquiry intention, and revise inquiry tactics , thus it can get the inquiry result as identical as user's actual demand. The relevant feedback system in this pper should be able to follow users' inquiry intention , judge which kind of methods is more suitable for the example image, every kind of method should play a same important role in first searching.Finally , we design and implement the prototype system with Visual C++.This prototype system can produce the color histogram and dominant hue histogram of a image, change the color space from RGB to HSI, and can choose arbitrarily image as the example images , then can search , follow user's inquire to the example image, verify the result of searching.