Localized region based face recognition using discrete krawtchouk moments

Abstract

Discrete Krawtchouk moments technique has been discovered as one of the feature extraction technique that is able to extract prominent features from the face and further improve face recognition performance. This technique is less sensitive towards image distortion due to its flexible characteristic in selecting a localized region of interest in face images by varying the value of Krawtchouk polynomial parameter. Therefore, this study intends to analyze the quality of features produced by a different value of Krawtchouk polynomial parameter and quantify using recognition rates produced on two different face databases: ORL and YALE. The study shows that face recognition using discrete Krawtchouk moments results up to 94.7% accuracy on ORL and 81.3% accuracy on Yale. Furthermore, this method outperformed two other classical feature extraction methods, known as Eigenfaces and Fisherfaces.