Virus Image Classification Based On Texture Analysis

Virus classification is an important research area in clinical virology. It is always a manual and time-consuming task, requiring highly skilled and experienced medical spe-cialists who perform visual examination using electron microscope (EM). Therefore, virus image classification method with high accuracy and automation has very important signif-icance in the field of clinical virology. Virus image shows a distinct and recurring texture. In recent years, some different virus image classification methods have been proposed, but the classification accuracies are not high enough. Therefore, in this paper, we study some classical texture analysis methods in recent years and propose the following two novel virus image classification algorithms:One of the algorithms is based on PCANet. Firstly, we preprocess the images in the data set. Then, the feature of preprocessed images are extracted by the stages of principal component analysis (PCA), binary hashing and block-wise histogram. Finally, the support vector machine (SVM) is utilized to do classification.Another algorithm is based on multi-scale PCA filtering and multi-scale completed local binary pattern (CLBP) feature extraction. Firstly, the filtered images are obtained by the multi-scale PCA filtering. Then, multi-scale CLBP feature is extracted from each filtered image. Finally, the SVM with polynomial kernel is used for classifying. The innovations of the method are extracting the virus image feature from the filtered images not the original image and applying the idea of "multi-scale" in the algorithm.The experiment results validate that, for the same virus image data sets, the classifica-tion accuracy of the algorithms proposed in this paper outperform the previous algorithms proposed in the existing literature.