Change detection by local illumination compensation using local binary pattern

Abstract

Approaches designed thus far for illumination invariant change detection are generally based on illumination compensation (IC) or illumination invariant descriptors. However, IC cannot handle local illumination variation, and is prone to sacrifice discriminability; illumination invariant descriptors cannot work very robustly in a textureless scenario. To address these problems, we present a novel change detection method by combining local binary pattern (LBP) with local illumination compensation (LIC). Although both LBP and LIC have disadvantages themselves, their independence enables mutual compensation of their disadvantages. Through a reasonable compositional strategy that makes best use of the advantages and bypasses the disadvantages, the proposed method can efficiently handle not only global but also local illumination variation in both textured and texture-less scenario. Experimental results using many real and synthetic images clearly justify our method.