According to the traditional AdaBoost algorithm with fast detection but low accuracy, a modified AdaBoost algorithm was presented to enhance the accuracy. First, the algorithm extracted Haar features of human face by rapid integral image. On the basis of this, it set the threshold value to modify the traditional AdaBoost algorithm and found the optimal weak classifier during each test, and then it cascaded them into strong classifier. Finally, strong classifier was developed to distinguish Haar feature and detect the part of face from images. The sample test results show that the classifier accuracy of FERET database is 96.07% and the video images is 96%. The experimental results demonstrate that the algorithm of human face detection designed can not only detect static images but also detect video images, which lay the foundation of face recognition and provide a kind of effective method for research of computer vision domain.