This paper proposes a novel foreground object extraction approach, which combines two kinds of feature information: image color and local image structure. Due to the highly complementary characteristic of the two used features, the proposed approach has a fairly good foreground object extraction ability when being applied to various situations. Compared with other commonly-used methods (such as color-based background extraction, temporal difference and optical flow), this method has shown two major advantages: (1) it can extract rather correct and complete object image even when the foreground objects and their background present similar colors; and (2) it performs much robust under the influence of lighting variation and shadow. Especially, this method is easy to implement and has a real-time execution performance. Consequentially, this method has a high practicability to various applications.
展开▼