基于医学图像降噪阈值选取的动态优化算法  

Medical imaging denoising threshold-based selection of dynamic optimization algorithm

作　　者：赵悟 段永璇[1] 肖宪翠 张睿[1] 席敏 ZHAO Wu;DUAN Yong-xuan;XIAO Xian-cui;ZHANG Rui;XI Min(Shandong Medical and Health Information Institute,Shandong No.1 Medical University or Shandong Academy of Medical Sciences,Jinan 250062,Shandong Province,China;Shandong Normal University Information Science and Engineering School,Jinan 250014,Shandong Province,China)

机构地区：[1]山东省医药卫生科技信息研究所,山东第一医科大学(山东省医学科学院),山东济南250062 [2]山东师范大学信息科学与工程学院,山东济南250014

出　　处：《中华医学图书情报杂志》2020年第4期72-76,共5页Chinese Journal of Medical Library and Information Science

基　　金：国家自然基金项目“网络资源协同管理与多目标虚拟网络映射问题研究”(61373149);山东省医学科学院面上项目“医学信息及数据服务云平台建设研究”(2016-01);山东省医药卫生科技发展计划项目“基于统计学习算法的精准疾病知识库构建研究”(2018WSA18030)。

摘　　要：SIFT算法是提取特征点的常用算法之一,具有良好的健壮性,在医学图像处理领域中广泛应用。然而传统的SIFT算法采用固定的降噪阈值会影响特征点选取的合理性,因此提出了一种动态优化算法(PM-SIFT算法),即采用PSO算法与互信息结合的方式优化传统SIFT算法,并从特征点有效性与计算时效性两方面对2种算法的特征点提取效果进行定量分析的结果表明,与传统的SIFT算法相比,PM-SIFT算法不但可以动态地选取合适的降噪阈值,而且还能提高图像配准的效果。Scale invariant feature transform( SIFT) algorithm,one of the commonly used algorithms for extraction of feature points,is of good robustness and widely applied in medical image processing. However,the invariant denoising threshold of traditional SIFT algorithm will affect the rationality to select its feature points,a dynamic optimization algorithm,PM-SIFT algorithm which optimizes the traditional SIFT algorithm by combining PSO algorithm with mutual information,was thus proposed in this paper. The feature pints extracted using the two algorithms were compared in terms of the availability of their feature points and the effectiveness of their computation. Quantitative analysis showed that the PM-SIFT algorithm can not only dynamically select appropriate medical image denoising threshold but also improve the effectiveness of medical image match as compared with the traditional SIFT algorithm.

关 键 词：SIFT算法 PSO算法 互信息 降噪阈值 特征点有效性 计算时效性 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]