Non-separable steerable filters for edge detection

Edge features can exist in an image at any possible direction and position. Thus, there is a need to find the image features at all possible orientations. This paper proposes a steerable filter based edge detection technique. The orientation tuning property of the steerable filters allows us to obtain filters at all orientations. Instead of using all orientations to get the final edge, limited number of oriented responses is used. To get the optimum response of all directions from limited oriented filters, two-dimensional non-separable filters´ responses combined to get the final edge detected image and qualitatively compared with that of the canny edge detector.