According to the problem of fast front face detection in complex background, a three cascaded fast front face detection algorithm. was proposed. At the first stage, non face regions were excluded through analysis of the maximum color connected regions using the HSV color model. During the second stage, the face areas were further accurately detected with Haar-like features and Adaboost classification algorithm. At last, an improved active shape model algorithm based on the local feature weighting (W-ASM) was proposed to match face feature points coordinates. Whether the current face image is front face can be judged by the locations of 68 feature points. The experimental results show that, this algorithm can accurately identify the front face perpendicular to the image rotating not more than ±5°. The average detection time of each image (640×480) is only 52ms, which can meet the real-time requirements.