Abstract: Tobacco leaf texture was different from tobacco leaf color and shape, it was more difficult to define and quantitatively analyze. Initially, it was concluded that tobacco leaf texture was related to the laxity of tobacco leaf surface tissue as well as folds and oil contents. The digital images of fresh tobacco leaves were obtained through an image acquisition system and texture features were extracted by the greyscale co-generation matrix （GLCM） algorithm and Gabor transformation. 112 dimensions of texture information were obtained with different gradient parameter settings, and the five to ten dimensions with the highest interpretation rate were obtained after principal element reduction and random forest classification as quantitative indicators of tobacco texture.

Key words: tobacco texture, texture feature extraction, principal component analysis dimension reduction, random forest