Study On Image Texture Feature Extraction And Classification

As a kind of fundamental property of the object’s surface, texture is widely used in imageanalysis, because it is a good characterization of the image. Texture analysis extracts importantgray information from the image with image processing techniques. The image texture featureextraction and texture classification are two important research topics in texture analysis. Thispaperemphasizesontheimageclassificationmethodsbasedontexturefeatureextraction. Usinga variety of texture feature extraction methods with some common classification algorithms, theimage classification achieves a good classification performance.The work done in image texture feature extraction and texture classification is mainly asfollows:This paper deseribes four major categories of texture analysis separately in detail，includ-ingstatisticsmethods,frequencyspectrummethods, modelingmethodsandstructuremeth-ods, and compares their advantages and disadvantages. In addition, it reviews some com-mon image classification methods, such as K Nearest Neighbour(KNN), K-means Clus-tering, Support Vector Machine(SVM), etc. and it points out strengths and limitations ofeach method.Aiming at the algorithm thoughts and implementation process in detail, this paper in-troduces the four common texture feature extraction methods, namely Gray Level Co-occurrence Matrix(GLCM), Tamura texture analysis, Gabor filters and Local Bianry Pat-tern(LBP).To achieve a multi-class texture classification, the K-means clustering algorithm which isa kind of unsupervised learning algorithm is selected, and adopts the Gabor transform toextract texture feature. Also, fuzzy K-means algorithm improved from K-means combinedwith GLCM to extract texture feature is used to classify texture image and compares itsmerits and demerits with the former method.Asanextendedapplicationofimageclassification, facerecognitionattractsgreatattention.BasedonSVMandtexturefeaturegeneratedbyGaborfilter, reducingdatadimensionwithPriciple Component Analysis(PCA), this paper analyzes the relationship between the face recognition and feature dimensions in feature space, and the relationship between the facerecognition and the number of training samples by testing on face image libray of ORLand YALE.At last, the presented work was summarized，and the further research and application oftexture features were discussed.