Abstract: The segmentation by C-V models can achieve good effect only if there are two smoothly partitioned zones in an image, i.e. the object zone and background zone. However, when extracting image boundary from complex background, C-V model often cannot get correct results. To solve the problem, the conventional C-V model and mean shift are used to develop a C-V model to adapt to the complex background. This model can determine piecewise the interior and exterior zones of contour curves. Experimental results showed that the model developed is more adaptive to complex background in comparison with the conventional C-V model and can extract smoothly uninterrupted image boundaries from complex background.