基于医学图像多特征的蚁群聚类检索技术研究  

Research of ant colony clustering optimization retrieval based on medical image multi-feature selection

作　　者：吕晓琪[1] 王新刚[1] 贾东征[1] 

机构地区：[1]内蒙古科技大学信息工程学院,内蒙古包头014010

出　　处：《计算机工程与设计》2014年第6期2078-2083,共6页Computer Engineering and Design

基　　金：国家自然科学基金项目(61179019);内蒙古科技大学创新基金项目(2012NCL018)

摘　　要：针对医学图像检索中单一特征无法有效表达复杂特征、易陷入局部最优的问题,提出一种将医学图像形状特征和纹理特征相结合的成对约束特征选择蚁群聚类算法。通过结合纹理特征灰度共生矩阵和形状特征Hu不变矩,建立多特征医学图像数据库集;用成对约束降维算法对医学图像数据集进行特征选择,形成最优特征集,对最优特征集进行动态加权,对加权最优特征集进行蚁群聚类;通过加权马氏距离相似性度量算法形成有规则的簇。基于改进的蚁群聚类医学图像检索算法相比传统的蚁群聚类检索算法,减少了专业人员对先验值的依赖,区分出了重要特征并按照重要的程度给予动态加权,该研究算法提高检索分类性能和稳定性,使检索查全率和查准率进一步优化,检索医学图像效果达到预期目标。To solve the problem that a single feature could not express complex medical image features and fell into local optimal in the image retrieval easily and to improve the existing medical image retrieval method,pairwise constraint method of semi-supervised learning features selection with an ant colony clustering method which combined shape features and texture features was proposed.Firstly,combining texture features of the gray level co-occurrence matrix and shape features of Hu moment invariant,a multi-feature of medical image database was set up,then appropriate medical image feature was chosen using pairwise constraint dimensionality reduction.Medical images with high similarity was clustered together by ant colony clustering algorithm.Finally,medical images in the dataset were retrieved by using similarity measuring algorithm of weighted Mahalanobis distance,which optimized the recall and precision ratio of retrieval and boosted the stability of classification and stability in the retrieval.The method,which combined the advantages of two features and optimized the effectivity of retrieval,did not depend on professionals prioris and overcame the problem that a single feature could not express complex medical image features,the expecting result was attained.

关 键 词：多特征 医学图像检索 蚁群聚类算法 HU不变矩 灰度共生矩阵 

分 类 号：TP311.5[自动化与计算机技术—计算机软件与理论]