Image Segmentation with Multi-Scale GVF Snake Model Based on B-Spline Wavelet

GVF snake model is insensitive to its initialization and can move into concave boundaries in image segmentation, however, it is expensive in computation and sensitive to noise. An image was decomposed to multi-scale images after the wavelet transform, then noise could be distinguished from signal by their different singularities in different resolution. In lower resolution, there were less wavelet coefficients, so the multi-scale GVF snake was easy to deform to the contour with less computation and robust to noise. In higher resolution, using initial contour yielded in the lower resolutions, the multi-scale GIF snake could get a finer result besides saving much more computation. The 3-order spline function was used as B-spline wavelet to implement the multi-scale transform. Experiments on MRI images show that the multi-scale GIF snake model is more quickly and more robust than GVF snake model.