Scaling laws in image gradient fields and texture retrieval

Under image scaling transform, two interaction models of image intensity gradient fields display unusual power scaling laws. These scalings are very universal in texture images and images of natural scenes, and robust against noise and image local deformation. An image similarity measure is developed in terms of these scaling laws. This similarity measure has graceful properties such as invariance under rotation, scaling and translation transformation, invariance under global image intensity adjustment and robustness against noise and image local deformation. A example of texture retrieval shows that the performance of this similarity measure is very promising, achieving 90% recognition rate in a test on a small image database