Edge detection in untextured and textured images-a common computational framework

In this paper, a unified computational framework is presented for facilitating edge detection both in untextured as well as textured two-dimensional (2-D) images. The framework is based on a complete set of difference operators which are easily configurable from a set of orthogonal polynomials. The widely known Roberts, Sobel, Prewitt, and Marr's LOG edge operators can easily be represented in terms of these operators. For detection of untextured or textured edges, the proposed operators are employed to separate out the responses toward edge or texture and noise. Untextured edges are detected by maximizing signal-to-noise ratio (SNR) or identifying the zero crossings in the second directional derivatives. Textured edges are detected in two stages. First, the significance of responses toward texture is computed statistically in order to test the presence of microtexture and compute a local descriptor called "pronum" for its representation. Finally, a global descriptor for texture called "prospectrum" is obtained by observing the frequency of occurrence of pronums. The textured edges are detected at the second stage by applying the methods of detection of untextured edges on these prospectrums. The results are encouraging.