REAL-TIME ON-LINE VIDEO OBJECT SEGMENTATION BASED ON MOTION DETECTION WITHOUT BACKGROUND CONSTRUCTION

A novel scheme for real-time on-line video object segmentation without background construction is presented. The proposed method uses foreground extraction-based video object segmentation. Motion and gradient-variant information is used to quickly acquire a coarse moving object mask. Compensation for still regions in a moving object is also proposed. Noise elimination, morphological processing and connected component labeling are used to obtain the fine moving object mask. Finally, moving object refinement (object boundary refinement, region growth/compensation and object region refinement) is used to overcome the residual background problem in order to obtain mole accurate video object segmentation. Experimental results show that the proposed method has good spatial accuracy, sensitivity, specificity and execution time. Objective evaluation results of the proposed method indicate that the average sensitivity, specificity and spatial accuracy can be maintained at 98.49%, 99.31% and 97.77%, respectively, for the tested video sequences.