Covariate shift approach for invariant texture classification

Abstract

This paper deals with rotation and scale invariant texture classification problem at the machine learning level by modelling these variations in the texture data as a covariate shift. Covariate shift between the training and testing data is minimised by estimating importance weights for the training data which are then incorporated in a standard machine learning algorithm like support vector machines. The effectiveness of these importance weighted support vector machines (IW-SVM) are tested on the Brodatz dataset. The comparative classification results with several other state of the art methodologies demonstrate the effectiveness of the proposed covariate shift approach for rotation and scale invariant texture classification.