An investigation of gabor PCA and different similarity measure techniques for image classification

Abstract

A method of attributing an image to a particular category from a large collection of images is stated as Image Classification. In this paper, we propose diverse subspace techniques, which concentrate on consistency and orientations of an image, extracting color, shape, and texture data with orthogonal transformation into uncorrelated space. Initially, preprocessing is done by transforming image to HSV color space as it is similar to human color perception property. Later, most informative score features are obtained using PCA, MPCA, KPCA, and GPCA with linear and nonlinear projection onto lower dimensional space which are further classified using diverse similarity measures and neural networks. The performance analysis is carried out on large multi-class datasets such as Corel-1K, Caltech-101, and Caltech-256 and the improvised correctness rate is witnessed in comparison with several benchmarking methods.