融合图像局部能量和梯度的水平集分割方法  

Image segmentation using a level set method based on local energy and gradient

作　　者：包立君[1] 刘宛予[1] 浦昭邦[1] 

机构地区：[1]哈尔滨工业大学HIT-INSA中法生物医学图像联合研究中心,哈尔滨150001

出　　处：《哈尔滨工业大学学报》2011年第3期44-48,共5页Journal of Harbin Institute of Technology

基　　金：国家自然科学基金资助项目(60777004);国家科技部重点国际合作项目(2007DFB30320)

摘　　要：为了实现结构复杂的灰度不均匀图像的快速准确分割,提出一种融合局部能量和梯度敏感性的水平集方法,采用梯度敏感的能量函数改进局部能量最小化水平集方法,并利用灰度全局信息自动初始化水平集.局部能量函数由局部灰度拟合函数定义,是水平集的外部驱动能量,适用于分割灰度不均匀图像.梯度敏感项依据图像特性,自动判定对水平集的驱动方向,其外部能量函数能加速零水平集向目标边界的运动,内部能量函数则推动零水平集离开平坦区域.该方法提高了水平集演化的速度和稳定性;通过调节水平集对不同强度边缘的敏感度,能够提取出弱边缘;而且不需要交互式操作.实验结果表明,该方法在分割灰度不均匀图像时,具有快速、准确和鲁棒性好的特点.To achieve quick and exact segmentations for intensity inhomogeneous images with complicated structures, we propose a novel level set method based on local region energy and gradient sensitivity, and initialize the level set automatically using the global intensity information. It is an improvement of minimization of region-scalable fitting energy with the introduction of two gradient sensitive energy functions. In this model, the local region energy based on the gray fitting functions acts as an external drive, which has an advantage in segmenting intensity inhomogeneous images. Gradient sensitive items can determine their drive directions of level sets adaptively according to image features. The external one draws zero level sets to the object boundaries, while the internal one pushes zero level sets out of intensity homogenous regions. Thereby, our method not only speeds up the convergence and improves robustness of level sets, but also copes with weak boundaries as the gradient sensitivity could be adjusted to respond to different gradient intensities. Meanwhile, interactive initialization is unnecessary. Experiment results demonstrate that our method has faster convergence, better robustness and good performance in segmenting intensity inhomogeneous images.

关 键 词：图像分割 水平集方法 局部拟合能量 梯度敏感性 自动初始化 

分 类 号：TN911[电子电信—通信与信息系统]