In chapter 4 , i performed the analyzing work with wavelet and fastica and find that wavelet was a high efficient way to filter the cep signal when i set the wavelet and scaling function to haar , decomposing level to 4 , threshold method to fixed to threshold and white - noise structure to scaled white - noise . it took the advantage that reduced the accumulating times to 60 with more smooth signal and less distortion . but fastica takes no advantage on this facet 本文通过小波变换和独立分量快速分析模式识别的方法对cep数据进行了分析处理，同时与数字低通椭圆滤波器的处理效果进行了比较，发现用小波对cep的数据进行滤波处理时，可以将数据的叠加次数从120降低到60次左右，且信号的质量要比以前的处理方法好；但是独立分量快速分析方法效果不很好。
Its basic thought is that using amass of simple classifier which has common classified ability and through thecertain method ， at last ， constitutes a very strong classifier which has strongclassified ability carries on many times with this strong classifier to the goalpictures ， finally confirmed the pictures whether includes the human face andits the general position this algorithm uses a characteristic which called haar characteristic thischaracteristic is one kind of simple rectangular characteristic ， because it issimilar with the haar wavelet ， so called haar characteristic this kind ofcharacteristic is composed of two or many rectangles that are congruent andneighboring there are white and black kinds of rectangles in the characteristictemplate , and defines this characteristic template characteristic value as thewhite rectangle this characteristic value is that the difference between white 本文主要研究基于haar特征的adaboost算法。由于以前提出的特征中包含的人脸基本特征比较少,导致检测时间过长。本文根据人脸基本特征的分布提出一种新的特征,新特征覆盖了人的眼睛,鼻子和嘴,它由haar特征中的一些简单特征组合而成,形状类似卷积中用到的3阶矩阵,这种新特征检测的结果是可以检测到人脸,时间上得到优化。
