Research On Image Retrieval Technology Based On Texture Spectrum Feature

Color, shape and texuture features are widly used in content-based image retrieval (CBIR). In this dissertation,the exploratory research work has been done around the texture feature extraction based on texture spectrum. The main contributions of this dissertation are summarized as follows:(1) Several key techniques and algorithms of CBIR are deeply analyzed and discussed, such as, the development and research content of CBIR, the low-level feature descriptions including color, shape, and texture,the evaluation methods of image retrieval algorithms. Especially, the decelopment and application of texture spectrum descriptors are discussed in detail.(2) A novel improved texture spectrum descriptor, called center-symmetric and center local binary pattern (CSC-LBP), is introduced. It uses the relation of 3 pixels in a neighborhood, the center and the center-symmetric pixels, to define the local texture patterns. On the other hand, the low frequency is also considered by comparing the average gray of the region and the whole image. The new operator fully uses the texture information contained in a region and produces a rather short histogram. It also has the same desirable properties as LBP, such as tolerance to illumination changes and computational simplicity. Based on the definition of rotation invariant texture (LBP Variance, LBPV), CSC-LBP is also extened to CSC-LBPV.(3) A novel texture spectrum descriptor, called direction local binary pattern (D-LBP), for region description is proposed. Because LBP produces a long histogram, the center-symmetric local binary pattern (CS-LBP) operator was introduced by Heikkil?. However, the center pixel of a region is ignored and it is hard to choose an adaptive threshold in this method. D-LBP considers the relativity of the center pixel and the pairs of the opposite pixels in a circle neighborhood. On the other hand, no thresholds are needed. The new descriptor is also improved and enhanced in the paper. Based on the definition of LBPV, the region texture spectrum are also extened.(4) The method of extracting spatial information of texute spectrum is reported. Based on the gray-level co-occurrence, the statics feature is extracted as the spatial feature of the texture spectrum. The methods mentioned in the paper are tested on three different databases and the results prove that the retrieval performance of the traditional texture spectrum descriptors could be improved markedly in image retrieval if the spatial feature is considered.