Pattern Recognition System Invariant to Rotation and Scale to Identify Color Images

摘要

This work presents a pattern recognition digital system based on nonlinear correlations. The correlation peak values given by the system were analyzed by the peak-to-correlation energy (PCE) metric to determine the optimal value of the non-linear coefficient kin the it-law. The system was tested with 18 different color images of butterflies; each image was rotated from 0° to 180° with increments of 1° and scaled ±25% with increments of 1% and to take advantage of the color property of the images the RGB model was employed. The boxplot statistical analysis of the mean with ±2~*EE (standard errors) for the PCE values set that the system invariant to rotation and scale has a confidence level at least of 95.4%.