采用Adaboost人脸检测算法, 对输入样本进行Haar特征提取, 从中选择出甄别效率最高的Haar特征, 然后将训练得到的Haar特征转换为弱分类器, 进一步将弱分类器组合成强分类器, 通过肤色确定候选点, 然后用强分类器进行人脸的检测, 进一步利用混合投影峰分析检测人眼的位置。通过C++实现了该算法, 完成了人脸检测算法的DSP程序移植。实验结果表明, 该DSP系统可以有效的实现人眼检测。

Adaboost face detection algorithm is applied in this practice. Firstly,it extracts the haar features of input samples and chooses optimal Haar features through training and changes it into weak classifer. Then,it makes weak classifier optimized into strong classifier. Candidate points are found by skin color features . Finally, faces are detected by cascade classifier and the location of eyes is detected by Hybrid Projection Peak Analysis. Adaboost face detection algorithm is implemented on C++,and the algorithm is transplanted to DSP. By the results, it is shown that this system on DSP can be used to realize face detection.