结合梯度和区域信息的多尺度水平集图像分割

Multi-scale level set image segmentation combined of gradient and region information

提出了一种结合梯度和区域信息的多尺度水平集图像分割算法。该算法结合梯度和区域信息构造能量函数,在梯度约束项中,构建了一个基于小波高频分量的边缘检测函数,在区域约束项中,运用经典C-V模型的区域项,得到混合C-V模型,采用变分法求解,并消除了水平集的重初始化。利用小波变换首先在逼近图像中运用混合C-V模型得到粗分辨图像的一个粗尺度分割,再对当前粗尺度下的最终轮廓线作内插操作,将得到的近似轮廓曲线作为初始水平集函数在原图像中运用消除重初始化的C-V模型演化得到最终的分割。实验结果表明,在同样的模型参数条件下,该方法具有比传统方法更高的演化效率和分割质量。

This paper proposes a multi-scale level set algorithm for image segmentation which combines of gradient and re-gion information.An energy function is constructed which combines of gradient and region information and gets a hybrid Chan-Vese model,which constructs an edge detection function based on wavelet high-frequency components in gradient con-straint term and applies region term of Chan-Vese model in region constraint term.Then use variational method to solve and eliminate the re-initialization procedure.The original image is firstly transformed into the wavelet domain to get a coarse ap-proximation,and an approximation contour is obtained on the coarse approximation by the hybrid Chan-Vese model.The ap-proximation contour is interpolated into the original-scale contour.Then the original-scale contour is taken as an initial level set function and the next active contour evolution which applies the Chan-Vese model of eliminating re-initialization is per-formed on the original image to get the real contour.Experimental results show that this method has higher evolution efficien-cy and quality than traditional methods in the condition of equal model parameters.