A new method for texture classification based on wavelet transform

The wavelet transform provides several important characteristics which can be used in a texture classification problem. We present a new algorithm based on conventional pyramid-structured wavelet transform. The iteration of the basic decomposition is submitted to a variance criterion which is applied on the current lowpass subimage. This criterion helps to decide the level where the decomposition have to be stopped. This level depends on the texture properties. It is used as a feature to assign textures into different classes. Form factors computed from wavelet decomposition are then used to perfect the classification process in each class. The performance of our method is pointed out on various images and compared with Laws's (1980) texture classification results.