Neural Classification of Objects Based on Gabor Signature

This paper uses a combination of K-Iterations Fast Learning Artificial Neural Network (KFLANN) and Gabor filters to create a Gabor signature classifier. Gabor filters are known to be useful in modeling responses of the receptive fields and the properties of simple cells in the visual cortex. The responses produced by Gabor filters produce good quantifiers of the visual content in any given image. A robust edge and edge orientation detection method using a combination of antisymmetric and symmetric Gabor filters is described in detail. The edge and edge orientation information are subsequently utilized to construct a Gabor signature that is size and orientation invariant. Some experimental results are provided to present the effectiveness and robustness of this signature construction for object classification. In addition to the KFLANN implementation, results were also obtained from a nearest neighbor classifier, back-propagation neural network and k-means clustering for the purposes of comparison.