Image-fusion-based contour extraction scheme

Abstract

A novel contour extraction scheme for detecting a moving target is proposed. This scheme consists of three steps. First, motion segmentation is applied respectively to infrared (IR) and visible image sequences to acquire an initial contour of the moving target. Second, dynamic contours are applied to make the initial contour converge to the target's contour with the Newmark-based iteration. Finally, two novel image fusions are applied to restrict the convergent dynamic contour in a visible image with that in an IR image. The first fusion minimizes the B-spline L 2 norm's square of the difference of control point vectors in the two modal images without image registration. The second fusion is realized by the revised differential coupling with image registration. A contrasting experiment on image sequence of a moving vehicle indicates that the contour extraction's average error decreases by 58.14% for the first fusion and 65.12% for the second fusion. Both image fusions are implemented with the control point vector of a dynamic contour and are suitable for practical application. Moreover, the Newmark based iteration is contrasted with the Wilson- θ -based iteration, which indicates that its iteration time requirement for convergence decreases by 21.01%.