Face contour extraction from front-view images

摘要

Facial feature detection is crucial for visual recognition of human faces. This paper reports on a study in detecting face contours from front-view ID-type pictures. Based on the eye and mouth positions, which can be detected by the algorithms proposed in Ref. 1 [G. Chow and X. Li, Pattern Recognition 26, 1739–1755 (1993)], a simplified adaptive Hough transform (AHT) technique is used to identify straight cheek lines which are approximately vertical from the edge image. Independently, parabolas forming the chin are detected by another AHT procedure. The location and curvature of the chin line, together with the cheek lines, characterize the shape of the face. This method was tested on over 70 different face images and is shown to produce results with a high degree of accuracy. We discuss in detail the specific issues involved in the detections, such as the definition of relevant subimage, parameter ranges, resolution of the accumulator array, peak cells, and end point determination.