一种自动初始化水平集的高频能量最小化三维分割模型

摘要

序列图像分割是医学图像三维重建的重要研究内容，但受成像技术限制，医学图像中往往包含大量低频信息，如偏移场、灰度不均等，影响分割准确性。从频域进行图像分割能有效避免低频信息干扰。在高频能量最小化分割模型基础上进行优化，设计了一种自动初始化水平集的分割模型并成功应用于三维分割领域。首先，使用形态学腐蚀方法进行粗分割，将提取出的三维曲面作为初始水平集，实现初始水平集轮廓面自动化；然后使用衍化后的水平集三维分割模型对其进行细分割。实验结果表明，采用该模型能够实现多目标分割，与原模型、Chan-Vese模型相比，分割结果更加准确。%Segmentation of images is an important research content of medical image 3D reconstruction.With the limitation of imaging technology,medical image often contains a lot of useless low frequency information,which leads to an inaccurate segmentation.To segment image in the frequency domain can effectively avoid the interference of low frequency informa-tion.With the ability to automatic initialize the level set function,the high frequency energy minimization segmentation model has been improved and optimized and applied to the field of 3D segmentation.Specifically,given an image sequence, first using the morphological method to get the initialized 3D surface,which is regarded as the initial 3D level set function. Then,refining the level set function with the proposed 3D segmentation model and get a more accurate segmentation re-sult.Experimental results show that the proposed model is able to achieve multiple obj ectives segmentation,and performs more accurate than the original model and the Chan-Vese model via excluding the influence of low interference.