ACM选择系统结合改进Chan-Vese模型的自适应图像分割方法  

Adaptive Image Segmentation Method Using ACM Selection System and Improved Chan-Vese Model

作　　者：何婷 郭艳光 辛春花 HE Ting;GUO Yanguang;XIN Chunhua(Department of Computer Technology and Information Management,Inner Mongolia Agricultural University,Baotou 014109,China)

机构地区：[1]内蒙古农业大学计算机技术与信息管理系

出　　处：《重庆理工大学学报（自然科学）》2020年第2期165-173,共9页Journal of Chongqing University of Technology：Natural Science

基　　金：内蒙古自治区自然科学基金项目(2015MS0565)

摘　　要：针对传统脑肿瘤人工分割方法稳定性与精确度不够高的问题,提出了一种ACM选择系统结合改进Chan-Vese模型的自适应图像分割方法。提出的框架适用于3种不同的局部区域主动轮廓模型(LRACM):LGDF、改进C-V以及LBF,根据要处理特定图像集,提出的方法可以自适应地选择其中最佳的一种模型来表示图像。首先,在学习阶段,其中一部分数据用于在最佳LRACM的选择任务中训练系统,并为此计算了平均值、调和平均值等10个图像特征;然后,在评估阶段,其余数据被测试以评估所提出的系统正确选择期望的主动轮廓模型的能力;最后,使用支持向量机分类器对3种模型所分割后的图像进行分类,将性能最好的模型作为所选模型,进一步进行准确分割。使用脑图像数据库Brain Perfusion Database的MRI数据和艾伦脑图像数据集进行实验。实验结果显示:相比仅单独使用其中1种模型的LRACM方法,提出的自适应选择方法实现了最佳分割效果。Aiming at the problem that the stability and accuracy of the traditional brain tumor segmentation method is not high enough,an adaptive selection of brain tumor magnetic resonance( MRI) image segmentation method combining ACM selection system and improved Chan-Vese model is proposed. The proposed LRACM can be divided into three models: LGDF,improved C-V and LBF. For this purpose,according to the specific image set to be processed,the best one of the models is adaptively selected to represent the image. First,in the learning phase,a part of the data is used to train the system in the selection task of the best LRACM,and 10 image features such as the average value and the harmonic mean value are calculated for this purpose. Then during the evaluation phase,the remaining data are tested to assess the ability of the proposed system to correctly select the desired active contour model. Using the support vector machine classifier to classify the images after the three models were divided,the best performance model was selected as the selected model,and further accurate segmentation was performed. Finally,experiments were performed using the MRI data of the brain image database Brain Perfusion Database and the Allen brain image data set. The results show that the proposed adaptive selection method achieves optimal segmentation compared to using a single LRACM method.

关 键 词：MRI图像分割 自适应选择 主动轮廓模型 支持向量机 CHAN-VESE模型 局部二值特征 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]