摘要 |
PROBLEM TO BE SOLVED: To increase learning efficiency without predetermining an optimum process and parameter of normalization. SOLUTION: For every combination of a plurality of "Rectangle Filters" and a plurality of normalization processes, the corresponding normalization process and "Rectangle Filter" are used to calculate evaluation values of all example images. Discrimination errors are calculated from the evaluation values. The combination of the normalization process and "Rectangle Filter" that cause the fewest discrimination errors is identified, and hypotheses are generated from the combination of the normalization process and "Rectangle Filter." All the hypotheses are combined to construct a discriminator. COPYRIGHT: (C)2007,JPO&INPIT
|