Improvement And Research Of Image Feature Extraction And Classification Algorithm Based On Tamura Texture Features

At present,as a universal way of information transmission,image has spread throughout all corners of human life and can be seen everywhere.Texture features,as an intrinsic property of images,play an important role in analyzing the information and connotation of images.At present,there are mainly four methods for extracting image texture features based on statistics,spectrum,model,structure and so on.All of them have various defects in varying degrees and can not fully meet people's daily needs.Firstly,this article makes a detailed explanation of three texture feature extraction techniques: Tamura texture features,gray level co-occurrence matrix,and local binary pattern.The Rosenfeld algorithm,which extracts the roughness of Tamura texture features,is introduced in detail,and Rosenfeld is improved.The calculation method of the domain mean difference in the algorithm and the quantization accuracy of the size quantization method are improved,the rotation invariance of the algorithm is effectively improved,the universality is greatly enhanced,and then the above original algorithm and the improved algorithm are respectively adopted.The Brodatz texture library and the Miao embroidery pattern were verified by simulation experiments.From the experimental results,it is basically consistent with the improved expectations.Finally,because most of image classification based on texture features is performed without prior knowledge,the affine propagation clustering algorithm suitable for this case is selected as the image classification method based on Tamura texture features.The similarity measure,bias parameter and attenuation factor setting have been studied accordingly,achieving the purpose of self-adaptation,which overcomes the defects of human control bias parameters and attenuation factors.Then,the affine propagation clustering algorithm is used to classify the images based on theTamura texture feature vectors extracted from Brodatz texture image database.According to the relevant experimental indicators,good classification results can be obtained(The Silhoutte index is 0.5831 under the characteristic distance),and the texture based on Tamura texture features is further verified.Feature extraction is feasible.