Abstract: Due to the success of sparse representation in face classification, recent researches on Sparse Representation-based pattern Classification（SRC） indicate the best sparse solution should be a research focus of face recognition. In order to do so and for the sake of the fact that the latest particle swarm algorithm called backbone particle swarm optimization（NPSO） has promising performance advantages over the conventional particle swarm algorithms, NPSO is adopted in this study. A new feature optimization algorithm called EnNPSO is proposed in which elastic network estimate is used to estimate the fitness function of NPSO algorithm and the pruning strategy is used to improve global exploration and stability. It enhances capability of handling high dimensional and small samples. Experimental results indicate the power of the proposed algorithm EnNPSO.

Key words: sparse representation, elastic network, face recognition, Particle Swarm Optimization（PSO）, bare bones Particle Swarm Optimization（PSO）, pruning strategy

摘要： 由于稀疏表示方法在人脸分类算法中的成功使用，基于此研究人员提出了一种新的分类方法即基于稀疏表示的分类方法（SRC）。因此寻求最优的稀疏表示方法就成为了人脸识别研究的重点。由于粒子群算法具有原理简单、参数较少和效率较高等优点，因此将基于剪枝策略的骨干粒子群算法（NPSO）应用于稀疏解的寻优过程。选择弹性网络估计（Elastic Network）作为NPSO算法的适应度函数，提出了一种稀疏解优化方法即EnNPSO。该方法具有很高的全局收敛性和稳定性，还具有很强的处理高维小样本和强相关性变量数据的能力。仿真实验表明该算法提高了人脸识别率，具有更高的适应性。

关键词: 稀疏表示, 弹性网络, 人脸识别, 粒子群算法, 骨干粒子群算法, 剪枝策略