Local binary pattern approach for rotation invariant texture classification

Abstract

In image processing, regular repetition of an element is known as texture. Texture classification is a process of assigning an unknown texture to a known texture class. In this paper, a simple, robust approach to texture classification is proposed. We develop a simple strategy to compute a local binary descriptor based on the conventional local binary pattern (LBP) approach, preserving the advantageous characteristics of uniform LBP. Points around center pixel are sampled such that for large circular neighborhood also it takes fix number of neighbors by local averaging. Then to make it rotation invariant we rotate the pattern circularly to a minimum number, leads to reduction in feature dimensionality.