融合纹理与形状特征的HEp-2细胞分类

摘要

间接免疫荧光(IIF)HEp-2细胞图像分析是自身免疫疾病诊断的重要依据,然而由于类内的变化与类间的相似性,HEp-2细胞染色模式分类具有很大难度.该文提出一种结合纹理和形状信息的有效分类方法,借鉴CLBP原理,提出具有完整信息描述能力的局部三值模式CLTP(Completed Local Triple Pattern)描述子来提取纹理信息,同时采用IFV(Improved Fisher Vector)模型和Rootsift特征来描绘形状信息,通过纹理和形状信息的结合,最终训练得到SVM分类器在ICPR 2012与ICIP 2013数据集上进行了对比试验.结果表明,所提方法在细胞级测试中优于其它方法,拥有竞争性的分类性能.%Indirect Immuno Fluorescence (IIF) HEp-2 cell image analysis is an important basis for the diagnosis of autoimmune diseases. However, due to the great changes in the class and the similarity between the categories, HEp-2 cell staining pattern classification is a difficult problem. This paper presents an effective classification method based on the texture and shape information, learning from the principle of CLBP, a descriptor extracting texture information is proposed to describe the Complete information of the Local Triple Pattern (CLTP). Moreover, using Improved Fisher Vector (IFV) model and Rootsift feature, the shape information can be described. Through the combination of the texture and shape information, an SVM classifier is finally trained and an experiment is conducted in ICPR 2012 and ICIP 2013 data sets. Experiment results show that this method is superior over other methods in the cell level test and present competitive performance.