Research On Content Based Classification Of Images

Along with the computer network technology and communication's fast development, the network learning resources present the multi-intermediaries and intellectualized trend of development,multimedia resources database and digital library construction become the time need. The image taking as an important media resources form, having the intuitive and vivid characteristic, is playing more and more important role in the educational and teaching practice, but because the existing resources database has not carried on the effective management and classification to the image resources, causing the user when sought for the resources to present the time-consuming and hard- sledding, actually not to find the needing resources. Therefore how to carry on the effective classification to the image resources, and how to bitterly serve the informational teaching, became the question urgently awaited to be solved.The present paper takes serving the educational resources database construction as a goal, and rests on the mentality of image content analysis, has carried on the digital image classification based on the content research. First through the observation and the investigation of 100,000 digital pictures, has established the image classification system which relying on the vision and the semantics. According to the level classification's method, we limited each kind of image's concept, analyzed the available visual characteristic, and provided the frame and the basis for the later image classification experiment. Next, the paper basing on the actual situation of the classified image, proposed the effective visual characteristic descriptor, including: the distance of color channel, the abundance of color clustering, the standard deviation of wavelet two decomposition, the fractal dimension and so on, and has introduced each descriptor's algorithm and the concrete application result. Once more, on the foundation of characteristic's extraction, we using the methods of threshold classification and SVM fuzzy classification, has experimented on the drawing/photography, line picture/ink picture, contour picture/sketch, Chinese painting/oil painting and so on, and has proposed 6 kind of available approaches, achieving the high accuracy and correct. Then analyzing each kind of classified experiment's result, pointing out the reason of wrong classification, we proposed the feasible improvement direction. We applying the result into the education video lens' classification, have carried on the close view/prospect classification experiment, and have obtained satisfying result. Finally, this article presenting the topic research by the form of website, we established"the image classification system based on the content", establishing 7 sector contents to demonstrate the classified system, the classified demonstration image and the research's new progress, and to facilitate with colleague's communication.This article based on the Windows XP operating system, taking VC++6.0 as the developing platform, unifying the Matlab 6.5 as data simulation tool, has developed"image classification based on content". Taking Dreamweaver 8.0 as the surface designing and the frame building tool, using the ASP as the procedure realizing language, has founded"Research on the construction of Digital image classification system". The image classification experiment has obtained the high rate of accuracy, and the website has achieved the good application.The findings indicated that the content classification of image not only liberated the manpower from the resources construction, but also avoided the resources redundant building. Therefore guarantying the objectivity of resources attribute labeling, is advantageous to the resources cross platform, trans-regional sharing, thus has provided the powerful guarantee for the educational resource's effective construction, better serving for our country's informational education.