This paper presents a new visual feature representation method as a weighted census transform (WCT) based on modified census transform (MCT) and entropy information of training dataset. The proposed feature representation model can offer robustness to represent same visual images such as MCT feature and sensitivity to effectively classify different visual images. In order to enhance the sensitivity of MCT feature, we designed the different weights for each MCT feature as binary code bit by statistical approach with the training dataset. In order to compare the proposed feature with MCT feature, we fixed classification method such as compressive sensing technique for two features. Experimental results shows that proposed WCT features have better classification performance than traditional MCT features for AR face datasets. identifier:https://dspace.jaist.ac.jp/dspace/handle/10119/11610