分形理论和参数活动轮廓模型分割医学图像

Fractal and Parametric Active Contour Model for Segmentation of Medical Image

针对传统参数活动轮廓模型对初始轮廓位置比较敏感的问题,提出了一种分形维数阈值法。对于不同的感兴趣区域选用合适的阈值,标识出分形维数满足该阈值的点,并用B-样条曲线拟合作为初始位置,实现了初始轮廓的自动定位。再利用GVFSnakes模型进行目标轮廓逼近,不仅大大减少迭代次数,还解决了传统参数活动轮廓模型凹陷边界收敛难的问题。实验证明了该方法的有效性,在信噪比较小的医学图像中分割效果较好。

A fraetal dimension threshold value method is introduced, since traditional parametric active contour model is sensitive to initial contour position. Appropriate choice of threshold value is made for different region of interest. B-spline is used to fit the points which meet the fractal dimension threshold. The target of the image is localized. Then the number of iterations is greatly reduced by the GVF Snakes model for the convergence to boundary contour. And the problem of poor convergence to boundary concavities of the traditional parametric active contour model can be solved. Experimental results show the effectiveness of the method. When the medical image signal to noise ratio is small, the effect is still good.