Combining geometric and gabor features for face recognition

Abstract

In automated face recognition, a human face can be described by several features, but very few of them are used in combination to improve discrimination ability. This paper demonstrates how different feature sets can be used to enhance discrimination for the purpose of face recognition. We have used geometrical features and Gabor features in combination for face recognition. The geometrical features include distances, areas, fuzzy membership values and evaluation values of the facial features namely eyes, eyebrows, nose and mouth. The Geometrie-Gabor features are extracted by applying the Gabor filters on the highly energized facial feature points on the face. These features are more robust to image variations caused by the imprecision of facial feature localization. An Extended-Geometric feature vector is constructed by combining both the feature sets and is found to achieve satisfactory results for face recognition using a simple matching function. The matching performance is analyzed for both the feature sets as well as for an Extended-Geometric feature vector. Experimental results demonstrate that no feature set alone is sufficient for recognition but the Extended-Geometric feature vector yields an improved recognition rate and speed at reduced computational cost and yet it is more discriminating and easy to discern from others. © Springer-Verlag Berlin Heidelberg 2006.