提取B超病灶的尺度分解小邻域统计水平集方法  

Multi-scale and small regional statistics level set method for lesion extraction in BU image

作　　者：杨谊[1] 喻德旷[1] 申洪[2] 高园园[1] 彭洁[1] 

机构地区：[1]南方医科大学生物医学工程学院,广东广州510515 [2]南方医科大学基础医学院,广东广州510515

出　　处：《南昌大学学报（工科版）》2014年第2期186-194,共9页Journal of Nanchang University(Engineering & Technology)

基　　金：广东省科技计划资助项目(2010B060300001);广东省产学研项目(2011B090400037)

摘　　要：在区域水平集方法基础上,利用小波变换将图像分解成多尺度子图像,对不同尺度子图像进行边缘提取并经过数学形态学算子修正,得到接近目标边界的水平集初始轮廓,解决了区域水平集方法对初始轮廓要求高的问题;接着运用小邻域统计方法计算并判断各像素点小邻域统计信息,筛选出该邻域内有效点集作为本次参数代入水平集公式进行迭代运算。实验表明:与几种现存的水平集方法相比,小邻域统计水平集方法能够较好地抑制B超图像噪声,快速地确定病灶边界,对模糊边界的捕获效果好。Based on regional level set method, the initial contour was get by decomposing the original image into multi-scale sub-images through wavelet transformation and obtains the edges, and get refined edges after mathematical morphology process as a good start for the level set contour evolution and meets the high demanding of the regional level set model. Then this paper makes use of statistics information of the mini-area centered with pixels on the contour during the iteration to filter the best valid pixels as the target pixels of each iteration. The small region statistic level set （SRSL） method is proved to be efficient in initial contour location and partially noise suppression, and can direct the contour to deform rapidly in intensity inhomogeneity and blurred areas, which results in better capture of weak edges of lesions than some popular level set methods.

关 键 词：水平集方法 尺度分解 小邻域统计 有效点集 B超病灶提取 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]