Moving object contour extraction based on change detection and improved GVF snake model

To deal with the problem of inaccuracy in current object contour extraction algorithms, a new approach based on improved gradient vector flow (GVF) snake and t-distribution significance test based change detection is proposed. It firstly eliminates most background boundaries through t-distribution significance test based change detection, an enclosed rectangle is obtained, and the rectangle is set to be the original boundary of the GVF snake model, which is obtained by calculating four critical points of the change detection mask. Then the improved GVF snake model is applied to the rough boundary to get the precise boundary. The improved GVF snake adopts anisotropic diffusion and a four-direction edge operator to solve the blurry boundary and edge shifting caused by the traditional GVF snake, and makes use of a new coefficient of fidelity term which has a faster descent speed to strengthen snake to segment the concave part. The proposed algorithm overcomes the disadvantages of drawing the initial contour manually. Furthermore, the GVF snake method in the spatial domain has been improved with higher accuracy. Experimental results indicate that the new proposed method gains accurate segmentation results in both the concave region and the weak edge part. Experimental results with several test video sequences indicate the validity and extraction accuracy.