Face detection and facial feature location is an important research in the field of pattern recognition. There may be glasses, beard, collar and other appendages as well as appearance, expression, color, lighting, shooting angles, all give rise to different images in face images. It brings a challenge to the face detection and facial feature location.This paper proposes some following improvements for the traditional Adaboost algorithm:1. Skin color segmentation. Comparing to several commonly color space in color images,the experiments show that it has a better effect in the separation of chrominance and luminance YCrCb space.Propses a new method that detectes the candidate face region.2. Because of the high false alarm rate of detction in complex background and the high time complexity of the Adboost algorithm. This paper presents a new face detection method: In the beginning,it uses the color detection to segment skin color in color space, then uses the Adaboost algorithm to detect the face, which greatly reduces the detection area of Adaboost algorithm and greatly improves the detection speed.3.The paper makes some adjustment to the update rules of the sample weights:Only the sample is incorrectly classified as well as its weight is less than the weight of the round update threshold, the weight of this sample will be increased, otherwise, their weight will be reduced. The experiments show that the adjustment of the weights has better effect in Adboost algorithm.4. This paper presents an improved algorithm for fast location of the human eyes: In the beginning,it uses the Otsu threshold, the vertical integral projection, the largest complexity to locate the human eye candidate regions,then uses the template matching to locate eyes precisely. This can greatly reduce the amount of computation. |