Autoregressive approach to surface texture analysis

The authors propose a major modification to the circular symmetric autoregressive random (CSAR) model by rendering the model rotationally invariant under all configurations of pixel implementation for texture analysis. It is based on the area segments covered by a circle set in a 3×3 neighborhood. It is found that the rotationally invariant property is true only for one condition: when the physical dimensions of the camera´s pixel constitute a square. Results obtained from the proposed circular area autoregressive (CAAR) model offer better consistency than that of the CSAR model for nonsquare pixel images. Hence, the CAAR model can be used to overcome the shortcomings of the CSAR model