The Research Of HEp-2 Cell Classification By Fusing Texture And Shape Features

Indirect immune fluorescence(IIF)HEp-2 image is an important basis for the diagnosis of autoimmune diseases.Immune fluorescence images are usually interpreted and analyzed by professionals.However,due to the limitations of subjectivity,different experts have the differences judges,thus affecting the reproducibility of results.At the same time,along with the popularization of indirect immune fluorescence image analysis,the demand for professionals is increasing,which greatly increases the labor intensity of professionals.In order to avoid the uncertainty of human diagnosis and subjective limitations,and reducing the labor intensity of experts,in recent years,the use of computer vision to achieve the diagnosis of widespread concern.In recent years,ICPR and ICIP have held the HEp-2staining pattern classification contest,sponsor can automatically compare IIF staining patterns of cells in the image,because the data set of public and evaluation standards,which greatly promoted the research and technical progress in the field.Because the feature extraction plays a key role in the staining pattern classification problem,it is more fully studied.At present,the features used are mainly divided into texture features and morphological features.Among them,the texture features are more widely used,and their performance is more stable,and the performance is better than the simple shape feature description.However,the existing literature methods often use a single texture or shape information,and very few of them apply texture or shape information together.Due to the imaging environment,the appearance of the same staining pattern will have a greater change,at the same time some pattern present very similar visual characteristics.For example,Homogeneous and Fine-speckled have very similar appearance,Fine-speckled mode and Coarse-speckled mode has very similar texture information,both of Nucleolar and Centromere have bright spots,only the spot size and number may be differences.The difference of some kind of staining patterns is mainly reflected in the appearance,while some difference is reflected in the texture,so if the texture and shape information can be combined,we believe the distinctiveness of the feature will be improved.This paper proposes an efficient classification method based on texture and shape information,using the principle of CLBP,the local three value model has theability to describe the complete information of the CLTP(Completed Local Triple Pattern)descriptor to extract texture information,while the use of IFV(Improved Fisher Vector)model and Rootsift feature to describe the shape information,through a combination of texture and shape information,we finally trained SVM classifier in ICPR 2012 and ICIP 2013 data sets,the test results show that this method is superior to other methods in cell level test,and present competitive performance.Compared with the best results of ICPR 2012 contest,our method improves the accuracy by about 7.7%.In addition,the method achieve 79.5% accuracy in ICIP2013 training dataset.