Application of Local Fisher Information Analysis to Salient Points Extraction

The Fisher information of local patterns about localization parameters can be used for selecting interesting points in computer vision applications. In this contribution it is shown that evaluation of local Fisher information about position, rotation and size for every point can be performed through a proper filter bank. It is recognized that this approach leads to an extension of the popular Harris salient point detection method, which actually searches for the patterns characterized by relative maxima of the Fisher information about translation. Preliminary experiments indicate that the Fisher information about rotational and scale parameters can be used to improve the repeatability rates of the salient point selection process with respect to the Harris and wavelet based salient point extraction methods.