Texture classification using multiresolution Markov random field models

摘要

Texture classification is an important topic in texture analysis. Texture classification has wide applications in remote sensing, computer vision, and image analysis. During the past years, several authors discussed to use multiresolution stochastic approaches to model textures. However, in these approaches, the highpass components which contain rich detailed information are lost. In this paper, we propose multiresolution MRF (MRMRF) modeling to describe textures. MRMRF modeling is a method trying to fuse filtering theory and MRF models. In the MRMRF modeling, highpass components are considered as well as lowpass components. "Brodatz texture database" is used in this paper for the experiments and Nearest Linear Combination (NLC) is used as measurement of distance to improve the recognition rate. The experimental results show that NLC has much better performance than Nearest Neighbor (NN) as the measurement in MRMRF modeling.