The invention provides a front face image detection method which includes the steps of capturing a video image, preprocessing the image by conducting median filtering, illumination compensation and image edge processing, detecting a face based on a method combining a cascade classifier and key characteristic points, and sorting out a front face image by using the geometric relation of the key characteristic points on a two dimensional plane. The method improves the image preprocessing method in a conventional face detection process. Interference on original image characteristics is reduced on the premises of realizing illumination compensation and de-noising. The error detection rate of face detection system based on an AdaBoost cascade classifier is lowered through the improved face detection method. Weight is added to a partial gray scale model of characteristic points through gauss distribution functions, thereby increasing the efficiency of a facial key characteristic points positioning method. A front face image is sorted out based on facial characteristic points accuracy positioning and by using the planar geometric relation of the characteristic points on a two-dimensional image.