局部二维结构描述的HEp-2染色模式分类  

HEp-2 Staining Pattern Classification by Local 2D Structure Description

作　　者：张东波[1,2] 颜霜[1] 张莹[1,2] 秦海[1] 王俊超[1] 

机构地区：[1]湘潭大学信息工程学院,湖南湘潭411105 [2]机器人视觉感知与控制技术国家工程实验室,湖南湘潭411105

出　　处：《电子学报》2016年第12期2817-2822,共6页Acta Electronica Sinica

基　　金：国家自然科学基金(No.60835004);湖南教育厅重点项目(No.14A137);湘潭大学控制科学与工程学科建设经费资助

摘　　要：Hep-2染色模式分类主要用于免疫疾病诊断,但已有方法受荧光成像环境,细胞图像自身的视觉特性的影响,分类准确率较低.本文提出一种新的适合于HEp-2染色模式分类的特征提取方法.在构建了不同尺度下的高斯平滑图像序列后,利用shape index实现图像二维结构的直观描述,进而通过多灰度阈值图像结构的空间分解,使其同时具备对微观二维图像结构和空间信息的描述能力.该方法在ICPR和SNP HEp-2数据集的两折交叉细胞级测试中,分别获得89.83%和87.49%的准确率,在ICPR的28折交叉细胞级和图像级测试分别达到60.5%和70.56%的准确率,明显优于LBP、CLBP等方法,和CoALBP特征相当.Hep-2 staining pattern classification is used for immune disease diagnosis. Due to the influence of imaging environment and the visual characteristics of cell images, the recognition accuracy of reported methods are still unsatisfied. A novel feature suitable for HEp-2 staining pattern classification is proposed. After constructing serial smoothing images with Gaussian scale parameter,the description ability of shape index is used to describe second-order image structure, moreover, the space structure can be presented by multi-threshold image segmentation. The proposed approach was tested in ICPR and SNPHEp-2 datasets by 2-rod cross validation in cell level, and achieved 89.83 % and 87, 49 % accuracy for the two datasets respectively, Moreover,in 28 fold cross validation test of ICPR, the method achieved 60. 5 % and 70. 56% accuracy in cell level and image level respectively. Experimental results show that the proposed method is superior to other popular texture descriptors, such as LBP and CLBP, and approximate to the performance of CoALBP feature.

关 键 词：间接免疫荧光图像 染色模式 空间分解 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]