A novel approach for computing the coefficient of ART descriptor using polar coordinates for gray-level and binary images

Abstract

ART (Angular Radial Transform) are widely used in many applications of domains of image processing and pattern recognition, this last is a region based shape descriptor adopted in MPEG-7; among the property, that descriptor is invariant to rotation. The direct computation of ART coefficients characterized by the basis function, which is defined in polar coordinates over a unit disk and digital images are usually defined in cartesian coordinates, which produces two types of errors namely geometric error and integral approximation error and invariant to rotation of ART is not verified. In this paper, we propose an algorithm to compute the coefficients in polar coordinates systems to eliminate these errors, which have been produced by the conventional method. We use cubic interpolation for generation of the pixel to system Polar image, the results show that the proposed approach improves the polar rotation invariance. In order to apply this approach at the gray level and the binary images, which we can compare the time of execution of this last.