A recent theory in vision science asserts that the information processing in early human vision can be modelled by a sequence of filters selective to certain bands of spatial frequency and orientation, In this paper, attempt to use computer alalogue the vision system. The Gaussian pyramid algorithm is used for spatial fequency filters. Selecting several feature-extraction operations of different orientations are regarded as masks, Identical multi-order statistics texture pairs generated by computer have been experimented according to Fisher linear diseriminant. Obtained classification acuracies are greater than 93% in the condition of better matching.