Feature Selection for Image Distortion Classification

Abstract

In this research, the spatial local mean subtraction contrast normalized coefficients are selected in order to achieve the classification of image distortion type. In order to achieve the feature selection, these coefficients and their products are calculated through four spatial orientations, which gives 18 features. Two methods for feature selection are applied in order to select a smaller number of features. Bhattacharyya distance is used as a measure that quantifies separability in the feature domain. The obtained results show that using methods for feature selection reduce the number of features and make the image distortion classification more robust.