A Gray Gradient Based Fast Training Algorithm for Face Detection

This paper describes a fast and simple training method for face detection based on block gradient of the gray level. The gradient orientation is one of the most essential features to describe the image structure. In this paper, the detector is trained just by the positive samples from which the feature values accumulated are regarded as the values´ weight to detect. The training set is of 50 faces simples. Each image is divided into three resolutions of 4×4, 8×8 and 16×16 to train the detector for different scales. While the detected value in low resolution is higher than the likelihood threshold, high resolution detector is used for a further detecting. Although the training framework is very simple, the correct detection rate is 91%.