Texture Image Classification Based On Markov Random Field

Texture is almost the inherent characteristics of all the image, and any image can be viewed as one or more texture combined. Nowadays, texture image analysis and classification is widely used in blemish locate, object recognize, image search, remote sensing image analysis and other fields. This dissertation has a study on the classification of texture images, It introduces the Markov Random Field (MRF) model and the method of feature Parameters selection, and then add some complementary features, lastly, BP artificial neural network is used to identify the texture images.Markov Random Field is extension of Markov Processing at the two-dimensional parameters set. MRF is probability model to express texture gathering in mathematics. The statistical parameters can express the magnitude and direction of pixels neighborhood, and describe the random characters of texture image. So MRF model is built according to the texture characteristics, and the texture parameters is acquired after estimation.With artificial neural network, the image texture according to MRF characteristic parameters. The neural network pattern-recognition is a new study direction of pattern-recognition in recent years. The result proves that MRF parameters is effective, the artificial neural network pattern-recognition is feasible.Besides, in this paper also added some texture features to express the characteristics, mainly including 8 characteristic parameters are extracted from the gray level co-Occurren matrix (GLCM), and the fractal dimension which can be express the texture feature information well. The experiments proved that ancillary features play a positive role for the classification of the texture images, and raise the recognition rate...