基于SWA的核自联想记忆模型及其人脸识别应用

Small-World Architecture Based Kernel Auto-Associative Memory Model and Its Application to Face Recognition

通过在传统的自联想记忆模型中引入机器学习中颇具影响力的核方法,提出了一类囊括现有自联想记忆模型的统一的核自联想记忆模型框架(KAM),并针对KAM所具有的复杂的全互连结构,借鉴最近由Watts和Strogatz提出的“小世界网络”理论,构建了一类结构相对简单、易于硬件实现的基于小世界体系(SWA)的核自联想记忆模型框架(SWA-KAM).在FERET人脸数据库上的随机加噪和部分遮挡的识别实验表明,该模型获得了比PCA算法以及最近提出的(PC)2A算法更高的识别率,表现出了较强的鲁棒性.

By introducing the kernel method into conventional auto-associative memory model （AM）, a unified framework of kernel auto-associative memory model （KAM） is established, which extends the existing AM. Taking into account the complex full connectivity of KAM, and based on the small-world network described by Watts and Strogatz, this paper proposes a framework of small-world architecture based kernel auto-associative memory model （SWA-KAM）, making VLSI implementation of AM easier. Simulation results on FERET face image database show that, SWA-KAM is more robust and has higher recognition rate than both PCA and （pC）^2A algorithms in the presence of additive noise or partial occluding on face images.