The second method of texture classification is the texton-based method. The specific method is to first extract the local features of the feature points, and then cluster to form texton; then calculate the statistical histogram for each frame as the final feature, and then use KNN or SVM for classification.
At the beginning, all prawns were made on the filter bank.ArticleThe method is similar to texton, but there are still differences. For details, see a statistical approach to texture classication fromSingle images. Well-known filter banks include: Leung-Malik (LM) set, Schmid (s) set, and maximum response (Mr or mr8) sets.
At the same time, the local feature points and description methods are also very popular. For more information, see local features and kernels for classification of texture and object.Categories: a comprehensive study. The methods for extracting feature points compared in this article include: Harris-Laplace detector and Laplacian detector. The descriptors include sift, spin, and rift.
Then there was an article about the trend in texture recognition: a statistical approach to material classification. Using Image Patch exemplars. This article uses the naked pixels around each pixel as the feature, and details why the effect is good, after that, follow is a previously submitted article that uses random projection for texture classification. For more information, see previous blogs: Http://blog.sina.com.cn/s/blog_631a4cc401013e4d. Html