基于改进活动轮廓模型和视觉特性的图像分割方法  

IMPROVED ACTIVE CONTOUR MODEL AND VISUAL ATTENTION BASED IMAGE SEGMENTATION METHOD

作　　者：付荣[1] 冉杨鋆[1] 孙晓光[1] 孙虎元[2] 孙立娟[2] 

机构地区：[1]复旦大学计算机科学与技术学院 [2]中国科学院海洋研究院

出　　处：《计算机应用与软件》2011年第9期5-8,33,共5页Computer Applications and Software

基　　金：国家自然科学基金(40776044)

摘　　要：提出一种基于改进活动轮廓模型和视觉显著性分析的图像分割方法。与传统的水平集模型不同,改进的活动轮廓模型不需要进行初始化和计算符号距离函数,从而有效地提高曲线演化效率。在此基础上,提出了基于标记的多相水平集分割方法,有效地解决了复杂图像存在的灰度不均性问题。另外,为避免初始轮廓位置对分割结果的影响,采用视觉显著图获取水平集初始轮廓位置,通过对该显著图进行OSTU分割提取初始轮廓。通过实验分析,提出的方法在分割结果、速度和复杂度上较之传统的CV模型都有明显的改进。An image segmentation method based on improved active contour model and visual attention analysis is proposed.Different from traditional level set models,the improved active contour model doesn′t require initializing or calculating the sign-distance function,thus effectively improves the curve evolutional efficiency.On this basis,a tag based variable level set segmentation method is proposed that effectively solves the problem of gray dishomogeneity in complicated images.In addition,to prevent the influence of the initial contour location on the segmentation result,the saliency map is utilized to obtain the level set initial contour location,which is segmented by OSTU to extract the initial contour.After experiment and analysis,the proposed method is proven to be obviously superior to traditional CV models on segmentation results,speed and complexity.

关 键 词：改进活动轮廓模型 视觉显著特性 水平集 图像分割 显著图 

分 类 号：TN911.73[电子电信—通信与信息系统]