Fisher Information Analysis for Matching Feature Extraction

The Cram-Rao inequality states that the reciprocal of the Fisher information is a lower bound on the variance of any unbiased estimator, which is used to the analysis of the object matching in the paper. Based on the Fisher information analysis, the lower variance bounds of the object matching transformation parameters are inversely proportional to the total gradient energy. So the pixel point gradient vector features are extracted for the machine vision object matching. And the mean value of the pixel point gradient normalized cross correlations is provided and used as the matching similarity measure. The pixel point gradient vector description of the object is more robust than image intensity, when there is scale variation, rotation variation or noise, and the object can be effectively recognized with the supposed matching methods, which has been verified by the experiments.