Cell Image Classification Based on the Support Vector Machine and D-S Evidence Theory

Abstract

When using the support vector machine (SVM) to directly classify the cell images according to their shape and texture features, the accuracy of that classification can be affected by any imbalance in the dimensions of these two types of features. We here proposed a method of cell image classification based on the SVM and D-S evidence theory. First, we performed "One-vs-rest" classification on the shape and texture features of the extracted images by using the v-SVM of the posterior probabilities. We then designed a reasonable reliability assignment function for the output probability classification results and carried out two rounds of classification of information fusion of the D-S evidence theory. Through the comparative experiments performed on the actual cell image data sets, we demonstrated that this method of design can classify cell images more accurately than other methods.