Feature Selection Method Based on Conditional Mutual Information for Image Classification
Shu‐Wei Jin · Ceshi jishu xuebao · 2010
A novel image feature selection method based on conditional mutual information was proposed.For predicting conditional mutual information,the method selected those features which has the maximum entropy with the features already picked,to be used for digital image classification.We implemented an image classifier by SVM.In the experiment,the input data of the classifier were grayscale images with the size of 28×28 and the 256 grayscale levels,and covered face and non-face images.We compared several image classification methods including the proposed method,the Bayesian theory,neural network and the kNN.The experimental results show that the method requires less time and higher correct recognition rate than other classifiers.