Abstract: Aiming at the limitations of the un-convex of energy functional in the active contour model, which makes the segmentation result sensitive to noise and the location of initial contours, a new model is proposed to make it be convex by the way of an un-constrained, and provides the fast algorithm based on the TV-norm. Experimental results on synthetic and medical images show that the proposed model can converges to a global minimum, accurately and fast segment the object, and is robust to image noise.

Key words: image segmentation, active contour model, level set, Total Variation(TV), energy functional

摘要： 几何主动轮廓模型的能量泛函是非凸性的，导致图像分割结果依赖于曲线的初始化条件，对噪声敏感。针对该问题，提出一种全局最小值分割模型，对能量泛函进行凸性非约束改进，利用基于总变分对偶公式的快速数值化算法实现图像的分割。对合成图像和医学图像的分割结果表明，利用该模型可以准确提取出对象的边界，分割速度快，对噪声具有较好的鲁棒性。

关键词: 图像分割, 主动轮廓模型, 水平集, 总变分, 能量泛函

CLC Number: 

• TP301.6