Optimal Weighting Functions for Feature Detection

One approach to feature detection is to match a parametric model of the feature to the image data. Naturally, the performance of such detectors is highly dependent upon the function used to measure the degree of fit between the feature model and the image data. In this paper, we first show how an existing detector can be extended to use a weighted L norm as the matching function with minimal extra computation. Next, we propose optimality criteria for the two fundamental aspects of feature detection performance: feature detection robustness and parameter estimation accuracy. We also show how to combine these criteria in various ways. We analyze the optimality criterion for parameter estimation under the approximating assumption that the feature manifold is locally linear. We also present a numerical algorithm that can be used to estimate the optimal weighting functions for the other optimality criteria. We include the results of applying this algorithm for step edge, line, and corner features.