Active Appearance Model (AAM) explains novel images by linear combination of statistic model which is build by applying Principal Component Analysis (PCA) on training data. The model of PCA essentially assumes that the observed variables, gathered in an uncorrelated way into vectors, but most of time it is not the case. In this paper, we present a probability method based on the analysis of the correlation between observed variables, which can better follow the distribution of data, to improve Principal Com- ponent Analysis which is used for feature extraction in Active Appearance Model. The major advantage of the proposed approach is proved to be more adaptive to the variable distribution. As consequence, images illuminated from different directions, and images with variable poses can easily be synthesized by changing the parameter found by KMCA. From the experimental result, the proposed method provides higher accuracy than classical Active Appearance Model for face alignment in a point-to-point error sense.