An anthropological model for automatic recognition of the male human face.

Abstract

The human face is a characteristic pattern most familiar to us when distinguishing people. Although recognizing human faces is one of our everyday activities, we are mostly not aware how the mechanisms of recognition actually work. Attempts to recognize the human face by machine are rarer (less frequent) than those of the recognition of some other phenomena in everyday life. This paper describes the automated analysis of a human face from the grey level picture, defined as an individual description of a face, given in the form of its sub-objects and the relations among them. The model-based analysis is founded on the anthropological model of a human face that incorporates 19 facial parameters of a male face in norma facialis. On the basis of these parameters it is possible to analyse, recognize and identify the human face. The contour image is used as an input to the pattern analysis program. Some algorithms for the search of characteristic face areas are presented. The Hough transform for an ellipse is used to determine the position of the head in a grey image, and to define the eye region within a face. The integral projections of the contour picture are proved to be useful techniques for picture processing to define the nose and the mouth regions within a face. Furthermore, a few algorithms for the precise determination of the facial parameters based on sharp edge transitions are developed. The method is illustrated using photographs of human faces from our data base and the results obtained are also given. The system is realized on a PC/AT computer.