Subpixel Edge Localization Based on Adaptive Weighting of Gradients

This paper presents a subpixel edge localization method based on the adaptive weighting of gradients (AWG). The method first finds the optimal power factors to be used by the AWG through an error propagation scheme. Then, with the goal of fast implementation of the method, a process that directly relates the edge width to the optimal powers is proposed. A method that uses the squared weighting of gradients (SWG) is also proposed, which applies a power of 2 to the gradients of an edge profile to calculate the edge location with subpixel accuracy. Next, an enhanced version of the AWG that combines the SWG and AWG selectively is proposed to obtain the best localization. At present, a fitting method based on the error function (Erf) is considered to be the most accurate method among current state-of-the-art methods, but it has a very high computational cost. The experimental results show that the proposed method is computationally much less expensive and more accurate than the Erf-fitting method.