基于稀疏表示和支持向量机的人脸识别算法

A Face Recognition Algorithm Based on Sparse Representation and Support Vector Machine

随着人脸识别技术的发展和应用,现今人脸识别的方法也趋于多样化,其中基于稀疏表示分类(SRC)的人脸识别方法是随着压缩感知理论兴起而诞生的一种全局线性方法。在先前研究的基础上,文中提出用正交匹配追踪法(OMP)代替梯度投影法(GPSR)来求解稀疏表示模型,通过设置稀疏阈值来控制稀疏系数的稀疏度,消除了非零系数出现在非样本所在类的现象。此外,基于SRC的人脸识别的识别准则是重构残差最小,对于一个测试样本,需要计算其与其他每一个样本的相似度,识别效率低。针对这个缺点,提出将多分类支持向量机作为最后分类的工具,在ORL人脸库上进行了实验验证,结果表明,该方法可以提高人脸识别的速度和准确率。

With the development and application of face recognition technique,the face recognition methods are also diversified at present.The face recognition method based on sparse representation classification （SRC） is a global linear method based on the rise of compressionperception theory. Based on the previous research,we propose solving the sparse representation model by orthogonal matching pursuit（OMP） instead of gradient projection for sparse reconstruction （GPSR）. The sparse threshold is set to control the sparsity of sparse coefficients,eliminating the phenomenon that nonzero coefficients appear in nonclass samples. In addition,the recognition criterion of face recognition based on SRC is the minimum reconstruction residuals. For a test sample,it is necessary to calculate its similarity to each other one andthe recognition efficiency is low. For this shortcoming,we propose a multi-class support vector machine as the final classification tool. Theresults on ORL show that this method can improve the speed and accuracy of face recognition.