Robust Estimation for Orientation Analysis

From trivial edge-detection techniques to the use of a quadrature pair of Gabor filters, the status quo in orientation analysis entails some form of convolution over a window of interest. Although Gabor filters have increased in popularity, most likely due to its computational efficiency as well as being a viable model for the human visual system, it is not without its shortcomings. Most prominent of those are the inconsistency of orientation estimates over a range of scales, the need for a large bank of differently tuned filters to account for all possible frequency patterns, and the inherent smoothing of high gradient regions resulting from its embedded Gaussian window. This paper proposes a practical alternative to standard Gabor-type approaches that allows for the presence of multiple orientations, offers increased robustness to noise while preserving such crucial scene elements as lines and edges.