Combination of Multiple Features for Object/Background Segmentation Using Graph Cut

Moving objects segmentation is a fundamental problem in many computer vision applications, which aims at detecting regions corresponding to moving objects such as vehicles and people in natural scenes. It is known to be a significant and difficult problem. Changes from illumination and shadow make segmentation difficult to process quickly and reliably. In this paper, a novel method is proposed to detect moving objects based on the linear combination of multiple features. Firstly, the color, gradient and texture features are synchronously used to construct background model, and the statistics of the background can be updated dynamically during processing. Because gradient and texture features are insensitive to illumination change, the method can improve accuracy of the objects segmentation. Secondly, graph cut algorithm is employed to compute the objects segmentation. Researchers have traditionally used combinations of morphological operations to remove the noise inherent in the result. Such techniques can effectively isolate foreground objects, but tend to lose fidelity around the borders of the segmentation, especially for noisy input. Graph cut algorithm results in qualitatively and quantitatively cleaner segmentation. Results can be temporally stabilized from frame to frame. The experimental results of different real scenes show that the proposed method is effective and practical.