Optical extraction of transform shape descriptors

Description of planar shapes is an important problem in pattern recognition. Hough transform (HT) is an efficient shape detector that maps straight lines into a two parameter feature space. Recently, it has been pointed out that the forward Radon transform (FRT), well-known from the theory of computed tomography, and the HT are equivalent for binary images. The FRT will not only be of use in implementing the HT shape descriptors but also to act as a coherent optical preprocessor for the implementation of multi-dimensional convolution, correlation and spectral analysis using 1-D acousto-optical signal processing devices. Experimental results, using conventional coherent Fourier transform configuration, are given. Fourier shape descriptors (FD's) have many properties; such as invariance to scaling, rotation and translation as well as the shape starting position. The FD's allow for considerable data reduction. Computer experiments show that the FD's are highly immune to noise and that correct shape reconstruction and classification can be expected with 7-9 db SNR's.