Hi, have tried to extract statistical descriptors, such as mean, variance, skewness and kurtosis? Local keypoint extraction might be a good method to detect some important informations (SIFT, SURF).
In this paper a new method for texture scale detection has been presented
SIFT descriptors also performed measures for orientation and scale information. SIFT keypoints are extracted in correspondence of texture, edge and high contrast regions.
I think this features can be used for texture classification.
Article Scale detection via keypoint density maps in regular or near...