Local Binary Pattern-based Noise Robust Feature for Texture Classification

Simarjot Kaur Randhawa; Ramesh Kumar Sunkaria

doi:10.37591/joipprp.v6i3.2206

Abstract

Abstract: The presence of noise degrades the local binary pattern-based classification efficiency. In the present work, a modified local binary pattern ( —modified noise robust local binary pattern) based classification is proposed. In this, a local binary pattern-based feature is modified, which also captures macrostructure information, whereas the existing features capture microstructure texture information only. The new feature is tested on Outex_TC_00010, Outex_TC_00012 and Brodatz datasets for rotation invariant and noise robust texture classification. The texture images are degraded with multiplicative noise, to evaluate noise robustness of the feature. Nearest neighbour classifier is used for classification which minimises chi-square distance. The proposed feature gives promising results as it is rotation invariant, robust to noise and gives high classification accuracy, especially at high levels of noise.Keywords: Texture classification, local binary pattern, feature extraction, histogramCite this Article: Simarjot Kaur Randhawa, Ramesh Kumar Sunkaria. Local Binary Pattern-based Noise Robust Feature for Texture Classification. Journal of Image Processing & Pattern Recognition Progress. 2019; 6(3): 31–47p.