Local Subspace Classifier with Gabor Filter Decomposition for Image Classification

This paper presents a local subspace classifier (LSC) with Gabor filter decomposition for image classification. In our method, first, the training images are decomposed into different directions by Gabor filters. By the same way as training images, an input image is decomposed into different directions with Gabor filters. After this, LSC is applied to each direction domain independently. The total sum of distances calculated from each direction is used for final classification. This method may be simple but we can improve accuracy by it. Experimental results on modified USPS, CIFAR-10, and SVHN datasets show that Gabor decomposition is effective for improving image classification accuracy of LSC.