多相水平集协同空间模糊聚类图像多目标分割

Multiphase Level Set-based Multi-objective Image Segmentation Cooperating with Spatial Fuzzy C-means

图像多目标分割的研究对于机器视觉发展具有重要意义。多相水平集模型(Multiphase level set,MLS)对零水平集函数初始位置和噪声敏感,当初始位置不适宜、噪声较大时无法准确分割多目标。针对上述问题,提出一种多相水平集模型协同空间模糊C-均值聚类(Spatial fuzzy C-means,SFCM)的图像多目标分割算法,即SFCM-MLS算法。首先用空间模糊聚类获取图像多目标粗分割结果,然后用粗分割结果定义多相水平集模型的初始水平集函数对图像做精分割。针对人脑磁共振成像(Magnetic resonance imaging,MRI)图像和患有肿瘤的肝脏计算机扫描断层图像多目标分割试验结果表明,与经典多相水平集模型相比,SFCM-MLS算法对初始位置不敏感,提高了图像多目标分割的准确性。

Multi-objective image segmentation is an important problem in machine vision. Multiphase level set model is sensitive to the initial contours of zero level set functions and image noise. Inappropriate initial position of zero level set or large image noise will lead to wrong segmentation results. A multiphase level set model-based multi-objective image segmentation method cooperating with spatial fuzzy C-means clustering, i.e. SFCM-MLS, is proposed. Coarse segmentation results are acquired with spatial fuzzy C-means clustering. Initial level set functions of muttiphase level set model are defined with these coarse results in order to finish the accurate segmentation. The proposed SFCM-MLS cooperating image segmentation algorithm is tested on magnetic resonance image of human brain and CT image of human liver with tumours. Experiment results indicate that compared with classical MLS, SFCM-MLS algorithm is insensitive to initial contours of zero level set and improves the accuracy of multi-obj ective image segmentation.