Performance Analysis of Local Binary Pattern Variants in Texture Classification

-Texture classification is a major issue in image analysis and pattern recognition. A number of methods are proposed in the literature including Local Binary Pattern (LBP). The LBP variant (s) plays an active role to extract texture features for texture classification. These are rotation invariant, noise sensitive or noise insensitive mehods. Each method has its own advantages and disadvantages. This paper is focused to provide a comparative analysis by evaluating the nine LBP variants using three well-known benchmark texture databases OUTEX, CUReT, UIUC using the nearestneighbourhood classifier. The nine LBP variants are rotation invariant and uniform LBP (LBP riu2 ), rotation invariant LBP (LBP ri ), Local Ternary Pattern (LTP), Variance (VAR), LBP and VAR (LBP/VAR), Completed Local Binary Pattern (CLBP), Completed Local Binary Count (CLBC), Adjacent Evaluation Completed Local Binary Pattern (AECLBP), Adjacent Evaluation Local Ternary Pattern (AELTP). The experimental results demonstrated that, Adjacent Evaluation Completed Local Binary Pattern (AECLBP) exhibits significant improvement in classification accuracy when compared to remaining LBP variants.