A Research Of Plate Texture Classification Algorithm Based On Gaussian Mixture Model

The plate texture recognition and classification is an important step in the process of wood plate production. The research on using computer vision technology to classify plate texture feature has great theoretical value and practical significance. The paper does some research based on the basic principles of local binary pattern (LBP) and gaussian mixture model (GMM). The main contents are as follows:The paper first proposes the idea of using LBP operator to describe the plate texture feature for the complex and ruleless plate texture structure on the basic principles of LBP conception, calculation method and correlation properties. It introduces the adjustment parameter ｓ to extend LBP operator for the purpose of enhancing the texture feature. At the same time, the paper combines the rotation invariance, uniform pattern with traditional LBP operator aimed at the shortcomings of large amount of feature values and classifying difficultly the similar textures before and after rotating image. The improved LBP operator is able to use less feature values to reflect more texture information in order to facilitate texture image recognition and classification.The paper proposes the idea of using GMM to classify plate texture feature and adopts EM algorithm to estimate model parameter on the basic principles of GMM conception, EM algorithm and the detailed derivation process of using EM algorithm to get GMM parameter. Because the initial value of parameter has an great effect on the result of EM algorithm, the paper adopts the result of K-means algorithm as initial value of GMM parameter. To some extent, this can improve the problem that initial values easily affect EM algorithm falling into local maximum, and can ensure the stability and accuracy of the results of classification.Through the above study, the paper combines LBP operator and GMM, and proposes the plate texture classification algorithm based on gaussian mixture model to realize the recognition and classification of multi-class plate texture. The experimental results show that the algorithm has the advantages of faster speed, higher accuracy of recognition and smaller storage space. Its effect is better than traditional gray level co-occurrence matrix, support vector machine, BP neural network and KNN algorithm. So the algorithm proposed in this paper has great practical value, and it provides a effective thought for the research on the plate texture classification algorithm.