Aiming at the high undetected rate and false detection rate, and other less which are existed in the AdaBoost algorithm based on Haar feature for face detection, the expanded categories of Haar features are added in this paper, and it can effectively reduce the erroneous judgement caused by the approximation of the gray value between the eyebrows and eyes. At the same time, the real-time of algorithm is improved by removing some features having bad effect for face detection. The cascade classifier constituting of Haar feature and AdaBoost algorithm is analyzed in depth. Finally, the experimental results verify the feasibility of the improved algorithm.