基于SVD和LDA的人脸识别方法

Face recognition based on SVD and LDA

提出了一种基于奇异值分解与改进的LDA相结合的人脸识别方法。首先利用奇异值分解方法获得图像的有效特征;然后经过改进的LDA处理,这样不仅可以有效降低维数,而且使抽取特征的判别能力得到了有效增强;最后对压缩后的特征向量进行排序,将排序后的特征送入BP网络进行识别。实验结果表明,该方法在低维特征向量下取得了很高的识别率,达到99%,效果优于传统方法。

This paper proposed a method of face recognition based on singular value decomposition and improved LDA. Firstly, effective feature could be obtained using singular value decomposition, and then improved LDA was used not only to depress the feature dimension effectively, but also to enhance the discriminatory power of extracted features. Finally, the short feature vector was sorted and the sorted features were input into the back-propagation network for recognition. Experimental results demonstrate that high recognition rate can be achieved using low dimensional feature vector and achieved 99%. This method outperforms traditional methods.