Dynamic Local Binary Pattern concatenation can be used for your objective. In this paper a solution is proposed. 

https://www.researchgate.net/profile/Guoying_Zhao/publication/6397809_Dynamic_texture_recognition_using_local_binary_patterns_with_an_application_to_facial_expressions/links/54b7c0940cf2e68eb2804972.pdf

Article Dynamic Texture Recognition Using Local Binary Patterns with...

hi please visit the link which take you to the university web page 

http://www.cse.oulu.fi/CMV/Research/LBP

http://www.cse.oulu.fi/CMV/Downloads/LBPMatlab

Thank's Mr. Bruno and Mr. Chathankumar

I would like to classify a large dataset of images, I have read in some articles that it divide each image into several sub-block and then extracted the texture improve the classification.

Is partition the image useful for classification process in the case of LBP?

I would say it may improve, especially if your image has strong structure.

I have also found useful to concatenate other types of features to the texture descriptor.

Thanks Dr. Henrique,

I have been studying some modification to LBP in order to discriminate more kind of texture. Some of them are based on making the LBP less sensitive to noise. Henrique what other feature you see usefull to concatenate with LBP histogram?.

What I meant was concatenating other texture descriptors like BGP or even LPB at a different scale, and also concatenating color features like an RGB or HSV miniature of the image in raster format. You can also play with PCA if you need a more compact combined descriptor.

There is always a trade-off between geometric invariance and having finer blocks over which LBP pattens can be computed and concatenated. If the texture patterns exhibit a lot of geometric transformations( rotations, scale, etc) beyond a point going for more detailed LBP may not help.