An Improved Texture-Based Method for Background Subtraction Using Local Binary Patterns

Abstract

Texture-based method (TBM) using local binary patterns (LBP) proposed in (1) is a successful solution to background subtraction especially for dynamic background scenes. However, it usually suffers from inaccuracy of the shapes of segmentation results and slow adaptation to the current situation. In this paper, we present an improved TBM that solves the two problems. To solve the first problem, a spatially weighted LBP histogram (SWLH) is proposed to be the feature vector and a simple shadow removing method is introduced. When dealing with the second one, we use an adaptive learning rate for each model LBP histogram and maintain multiple frame level models to process sudden illumination changes. Experimental results show that the proposed method outperforms the original TBM.