多尺度多特征仿生人脸识别

Multi-scale and multi-feature bionic face recognition

本文使用Daubechies正交小波变换对人脸图像进行二次小波分解:首先对第二次小波变换低频子图像进行PCA分析。运用邻域法进行分类得到距离隶属度。利用模糊分析提取出候选样本,对候选样本第一次小波变换的低频子图像进行PCA分析,运用最近邻域法进行分类得到最终识别结果。实验表明:小波变换预处理得到多尺度多特征;分类结果之间具有一定的互补性,同时可以提高分类性能。

Danbechies orthogonal wavelet transfonn is applied to preprocess the face image. Firstly, principal component analysis （PCA） is used to low resolution sub- image which is generated by the second wavelet transform. The neighbor classifier is applied to get the distance feudatory. Fuzzy decomposition is employed to extract candidate samples. PCA is used to analysis the low resolution sub - image, which is gained by first wavelet transform. The nearest neighbor classifter is applied to recognize them. Experiments show that the multi - resolution sub - images and multi - feature get by the wavelet transform and that so far as the PCA analysis is concerned, there is a certain complementarily between the sort results of the sub - images, The sort performance can be improved.