基于图像分块的二维投影特征提取能力研究

2-Dimension Projection Feature Extracting Ability Research Based-On Image Module

人脸识别研究的主要目的是提高识别率,关键技术在于提取有效的人脸特征。提出了分块多投影和分块双向多投影二维特征提取方法。分块多投影特征提取方法,针对现有分块单投影特征提取方法中每一子图均采用相同投影矩阵,而对人脸局部信息不加以区别,利用二维主成分分析方法,构造了分块多投影矩阵,使不同的子图投影到不同的子空间,与传统二维主成分方法和分块单投影方法相比,有效地利用人脸局部信息,降低了特征维数,提高了识别率,在ORL人脸库上实验表明了其有效性。

The target of face recognition is to enhance recognition correctness. The key technology is how to ex- tract effective features of face characteristics. This paper proposes the 2 - dimension feature extraction methods of modular multi - projection and modular two direction multi - projection. Targeting the problem that all the sub - ima- ges choose the same projection matrix and don＇t discriminate facial local information in modular single - projection feature extraction method, module multi - projection feature extraction method constructs modular multi - project ma- trix using the traditional 2 - dimaension principle component, making different sub - images to project different sub- spaces. Compared with the traditional 2 - dimaension principle component and modular single - projection feature ex- traction method, recognition rate of this method is greatly improved and the dimension of feature is reduced effectively by making use of facial local information. Experiments on ORL face database proved its effectiveness .