For each picture a signature is generated by operating on it with different masks. The operations are a generalization to grey level images of 'shrink' and 'expand' of binary pictures using J. Serra's morphological methods. The signatures are sets of numbers, and can be used to analyze and discriminate textures. It is shown that this family of operators includes as private cases several currently used texture descriptors. A metric between matrices is also proposed that improved classification over ordinary distance.