Low-Cost Junction Characterization using Polar Averaging Filters

Junction characterization is a very costly task since junctions are multi-scalar and multi-oriented structures. Besides, high directional selectivity is also required to deal with complex junctions. Steerable lters attenuate this computational burden by approximating a lter at an arbitrary scale and orientation with a-nite set of so-called basis lters. But due to the uncertainty principle high orientational selectivity can be only achieved by applying a large number of basis l-ters. In this paper we present a kind of lter with small spatial support to alleviate this eeort. The mask of our lter is determined not by Fourier transform, but directly according to the coordinates. Hence, masks with high orientational selectivity can be obtained without increasing any extra computation. Moreover, only averaging values in the masks are calculated. The expensive 2D convolution of normal ltering is therefore avoided.