Abstract

Many techniques have been proposed for edge detection involving transforms, such as the method of Shanmugam et al.¹ and the gradient method of Marr and Hildreth.² It can be easily shown that such methods are some kind of bandpass filtering. Because of the existence of different kind of edges and different amount of noise in a real image, no unique filter can be optimal. We discuss how to use some novel real fast transforms for edge detection through bandpass filtering. These are discrete cosine transform (DCT), real discrete Fourier transform (RDFT), scrambled real discrete Fourier transform (SRDFT), and discrete cosine-III transform (DC3T). These transforms also show little block effects, compared to the discrete Fourier transform. They can also be used for interpolation to increase the resolution of edge location and decrease the effect of inherent noise in the real image. For both bandpass filtering and interpolation we applied transforms blockwise to decrease computational complexity to achieve parallel implementation.

© 1991 Optical Society of America