Rotation and scale invariant texture classification by compensating for distribution changes using covariate shift in uniform local binary patterns

A novel rotation and scale invariant texture classification methodologyis proposed based on distribution matching in higher dimensionalspace. Feature extraction is performed by using uniform local binarypatterns (uLBPs) in which the rotation and scale changes in animage cause shifts in the underlying uLBP histograms. To compensatefor these shifts at the classification layer, the distributions of trainingand testing data using kernel methods are estimated and means ofthe two distributions in the transformed domain using importanceweights are matched. These calculated importance weights are usedin the standard support vector machines to compensate for the shiftin the distributions. The proposed method is used for classifying theimages in the Brodatz texture database demonstrating the effectivenessof the proposed methodology.