基于多尺度图像局部结构分解的人脸特征提取方法

Face Feature Extraction and Recognition Method for Multi-scale Local Structure-based Image Decomposition

为了有效提取人脸图像的全局和局部特征以提高人脸识别的性能,提出一种基于多尺度图像局部结构分解的人脸特征提取方法。该方法首先通过多尺度分析构建人脸图像金字塔,然后对于金字塔中每一层的图像应用脊回归度量图像局部窗口内中心宏像素与其近邻宏像素之间的结构关系从而刻画出图像的局部结构信息,再根据得到的局部结构信息将图像分解为若干个子图像,最后将这些子图像均匀下采样和归一化后连接在一起形成一个特征向量。实验结果表明,与Gabor、LBP和IDLS等方法相比,该方法具有更好的识别性能。

In order to effectively extract the global and local features to improve the performance of face recognition,this paper presents a robust yet simple feature extraction method,called multi-scale image decomposition based on local structure. In the algorithm,the face image pyramid is first constructed through a multi-scale analysis. Then the local structural information by describing the relationship between the central macro-pixel and its neighbors for each level of the image pyramid is captured. In this way,one image is actually decomposed into a series of sub-images. Finally,all the structure images,after being down-sampled,are concatenated in one super-vector. Experimental results show that the proposed method is superior to some traditional methods such as Gabor,LBP and IDLS.