Research On Classification And Retrieval Of Texture Images

With the development of the Internet, which is progressing in high speed towardswide-band and multi-media, it is providing us with more and more available resources,such as the text, the image, video and audio resources and so on. As an important partof the network information, educational resources play a significant role in theimprovement of the educational quality and the realization of its potentials. Among allkinds of information resources, images are much more concrete and intelligible,delivering generous information. Therefore, it has become one of the most significantparts that constitute the educational resource library.Just like all the other network information, the enormous number, the variety andcomplex sequences of the educational images obstruct the advancement of the imageretrieval to a large extent. In recent years, the content-based image retrieval hasimproved rapidly. It's mainly based on the content and describes the reasonablefeatures of the images in order to make the retrieval more efficient and adjust topeople's vision as satisfying as possible. The content of images is made up with color,texture, shape , language features and so on. Among all the above, texture is one ofthe most remarkable features. However, texture features haven't been made full use ofin the content-based image retrieval yet, which description and analysis are complex.In this thesis, we classify the natural texture into ten classes according to theirconception. In addition, we study the texture segment and texture-based imageretrieval of natural images.We apply the texture conception of natural language to the texture classification,and classify the natural texture into ten classes that are YuLin, Keli, Liewen, Banwen,Tiaowen, Rongmao, Bowen, Muwen, Huawen and Luanwen. Basing on the above, wefound a small image library of natural texture. In the thesis, we discuss the commonmeans of texture feature extraction, analyze the Wavelet Packet, and bring forward aspecific algorithm for Gabor filter. In order to verify the validity of the featureextraction, we adopt the BP network and SVM as the classifier to carry out ourexperiments, which bring us satisfying results.Texture segment is an essential step for texture image analysis. Only after segment,we can extract the visible and language features. Here we mostly study the GrayLevel Co-occurrence Matrix. We research into the influence of the distance, the graylevel and the window on the segment result. In the thesis, we make use of the fuzzy ccluster to classifier different texture, and get satisfactory results, too.Finally, we discuss the features of the content-based image retrieval, carry on atexture-based image retrieval experiment, and get a satisfactory result.The whole test platform is based on the Microsoft Windows 2000 System andAccess database system. Using the Visual C++ 6.0 and Matlab 6.5, we explore anatural texture classification and segment system and a texture-based image retrievalsystem. The experiment results indicate that our algorithm is effective.Our research proves that making use of the texture features in content-basedretrieval can improve the precision of the retrieval to a large extent. Therefore we canlet the network images serve the building of our Education Resource Library.