基于Alpha散度的NMF人脸识别方法

传统的人脸识别方法对图像质量要求较高,对含有噪声或复杂背景等真实世界的图像识别率较低,从而限制了人脸识别的应用.基于Alpha散度的NMF分解方法用于人脸识别,用Alpha散度作为距离度量标准,得到对应的NMF分解表达式,通过表达式中参数的取值可以衍生出多种分解迭代表达式,在每步迭代过程中计算差异度,进而确定下一步的最优参数,这样能保证分解收敛于全局最优,提高人脸识别的精度.

An NMF Face Recognition Method Based on Alpha Divergence

The traditional face recognition methods require a relatively high quality image,as such their recognition rates are comparably low for some real-world images that contain noise or have complex backgrounds.As a result,applications of face recognition are limited.This paper proposes an NMF decomposition method based on alpha divergence for face recognition,which uses alpha divergence as distance measure standard to obtain the corresponding NMF decomposition expressions.Through the parameter values derived from the expressions,a variety of decomposition iteration expressions can be obtained.In each iteration process,the differences are calculated to determine the optimal parameters of the next step.Such decomposition can converge to the global optimum to improve the accuracy of face recognition.