Research On Texture Feature Extraction Methods Of Color Image

As a classical problem of computer vision,texture classification has received extensive attention from scholars at home and abroad.Texture feature extraction is the basis and prerequisite for texture classification.However,most texture feature extraction methods are proposed on the basis of grayscale images,ignoring color information.The real word is colorful,and color information is also essential as a discriminative feature.For the extraction of color features,most methods are to directly combine the features in separately extracted from every of three channels.Since each color channel is not independent and thus there is a certain relationship with each other,it is more important to capture the characteristics between different color channels.In view of above problems,this dissertation presents three color texture feature extraction methods,and the texture classification experiments based on three feature extraction methods were conducted on the published standard color texture datasets(CURe T ? Vis Tex ? Colored Brodatz ? USPTex ? KTH-TIPS).Experiment results show that the presented methods are superior to other representative color feature extraction methods.Specifically,the main innovations of this dissertation include the following three aspects:1.A color texture feature extraction method based on the shortest path in HSI space is proposed.By modeling the H color channel of the color image as an undirected graph in the HSI color space,the S color channel and the I color channel are jointly modeled as an undirected graph,and two undirected graphs are used to represent the color texture image.Finally,the first-order statistics of the shortest path in four directions in the undirected graph is calculated to construct the color texture feature.2.A color texture feature extraction method based on compact interchannel sampling difference is proposed.Firstly,the t distribution is used to sample the microblock in the image patch,and then the dense micro-blocks difference is used to model the relationship between the different color channels of the color image to capture interchannel difference features.Finally,the Fisher vector is used to encoded the obtained difference features,and PCA is used to reduce the dimension of the encoded features to obtain a compact interchannel sampling difference descriptor.3.A color texture feature extraction method based on completed extremely nonnegative DMD is proposed.Firstly,the DMD is used to model the intrachannel features and interchannel features of the color image.Negative values are meaningledd in digital images,so the nonnegative operation is used in the difference process.The extremely nonnegative DMD is constructed by fusing the maximum valued of the three intrachannel features and the three interchannel features,the extremely nonnegative DMD including local features and the global feature of the color texture are fused to construct a completed extremely nonnegative DMD.