Information properties of Fourier-Mellin image descriptors
Yajun Li, Francis T. S. Yu · Optical Society of America Annual Meeting · 1990
The Fourier-Mellin image descriptors (FMDs) are invariant to both a scale and a rotation change in the input image. The FMDs can be calculated by using a computer, but the computational load is heavy.1 In this paper, we have proposed a method for calculating the accurate values of the FMDs of binary patterns constructed by rectangular pixels. This method was successfully applied to the calculation of the FMDs of all the alphanumeric characters that are more complicated patterns than squares, square frames, and rectangles. The results clearly show the information properties of the FMDs, some do not provide enough information for discrimination between characters and others just emphasize the same parts. It is known1,2 that the values of the high-order FMDs are dominated by the outer portion of a visual pattern and that the radial information from the central portion is suppressed. A suggestion has been made to reduce the unbalance in information extraction.