Invariant feature extraction with exponential chirp transform

Abstract

To recognize an object with an arbitrary attitude, the employed algorithm must guarantee an ability to extract invariant features against various attitudes of the target object. In this paper, we propose a new method to discriminate an object in 2D image plane moved with unknown parameters of motions such as rotation, dilation, and translation. The method is based on the exponential chirp transform(ECT), proposed by G. Bonmassar and E.L. Schwartz, to perform the complex-logarithmic mapping(CLM)with a slight offset for avoiding the singularity around the origin. There is a fast algorithm for ECT, and the size of images to be processed is considerably reduced by the CLM, thus yielding a fast performance for pattern recognition. Experimental results show that an object with arbitrary motions is reliably discriminated.