Facial Expression Recognition Based on Gabor Filter Performance Analysis

Abstract

This paper extracts the Gabor phase feature information to classify facial expression. First, preprocessing the image for obtaining the normalization image of pure expression, Gabor transform has good space-frequency localized and multi-directional selectivity, so uses Gabor filter with five frequencies and eight directions to filter the pure expression image. By changing the filter's center frequency, get the optimal image after filtering, and then extract the phase features, carry on the dimension reduction. Finally, with nearest neighbor classifier to classify, a better experimental result had shown in JAFFE database.