基于改进Adaboost人脸检测算法的研究及DSP实现

在Adaboost人脸检测算法的基础上,通过将所有的特征值排序并构建一个次序表,并使用改进Adaboost权值更新算法和分类误差计算方法,训练分类器并检测样本图像,达到了减少训练分类器耗时、提高检测率和降低误检率的效果.最后在DSP上实现了人脸检测系统,验证了改进系统在实时性能上得到较大的改善.

Research and DSP Realization on Face Detection Based on Improved Adaboost Algorithm

In order to solve the problem that the training time of Adaboost face detection is extremely long,and to raise the detection rate while lowering the false detection rate,several improvements based on Adaboo1st face detection algorithm are used to train and detect the faces from the importing pictures.Improvements included in this thesis are to build and store a feature form in a decending order,to change the weight renewal algorithm and classification error calculation method.At last,this improved method is realized on DSP platform.It is proved that the real-time feature is much better than that of the normal Adaboost.