Object Recognition from Color Images by Fuzzy Classification of Gabor Wavelet Features

Abstract

This paper introduces a novel object recognition approach based on the Gabor Wavelet representation of the binarized image that makes use of fuzzy logic for determining the 'soft' class label of the color test images with respect to the gray training templates. The fuzzy membership function used is a Generalized Gaussian function whose exponent value is determined empirically. The use of simple computations for assigning the class label to the query image makes the technique computationally effective. The experimental results on 494 color test images from ten object categories from the Caltech database show a high percentage of classification accuracy with only fifteen gray images from each category used for training. The efficiency of our method is established by comparing our results with that of different classifiers and also with Qiu's Content based Image retrieval (CBIR) system for color images.