Feature value smoothing as an aid in texture analysis

Abstract : When texture features are measured on small subimages, they are unreliable; but if we use large subimages, it is hard to find subimages that are uniformly textured. This paper describes a compromise approach: measure the features on small subimages, and smooth the resulting feature values in such a way that neighboring subimages that belong to differently textured regions are unlikely to influence one another. When this is done, classification performance improves substantially. Improvement is also obtained when the subimages are classified probabilistically and relaxation is used to adjust the class probabilities. The problem of choosing a window size that minimizes overall misclassification probability is also discussed.

[1]  Patrick C. Chen,et al.  Segmentation by texture using a co-occurrence matrix and a split-and-merge algorithm☆ , 1979 .

[2]  R.M. Haralick,et al.  Statistical and structural approaches to texture , 1979, Proceedings of the IEEE.

[3]  Azriel Rosenfeld,et al.  Scene Labeling by Relaxation Operations , 1976, IEEE Transactions on Systems, Man, and Cybernetics.

[4]  Shmuel Peleg,et al.  A New Probabilistic Relaxation Scheme , 1980, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[5]  Azriel Rosenfeld,et al.  A Relaxation Method for Multispectral Pixel Classification , 1980, IEEE Transactions on Pattern Analysis and Machine Intelligence.