Improved Multi-spiral Local Binary Pattern in Texture Recognition

Abstract

Local Binary Pattern (LBP) is a well-known appearance-based local feature descriptor. Since it is successfully applied to many pattern recognition applications such as texture recognition, face recognition, and so on, many variants of LBP are proposed by researchers. It is known that edges carry important discriminative information about the geometric structure and content of the image. In this paper, the discrimination ability of the feature descriptors derived from the edges of an image using Spiral Local Binary Patterns (S1BLP) and its two variants, namely two Spiral LBP (S2LBP) and four Spiral LBP (S4LBP) are investigated. We also combine this descriptor with S1LBP, S2LBP, and S4LBP features which are derived from the whole image. Linear Regression Classification (LRC) and test are used to investigate the performance of the proposed descriptors in terms of classification accuracy. The classification tests conducted on two different texture datasets, namely CURet and UIUC show that the proposed feature descriptor has important discriminative information which improves the classification accuracy.