基于CSSD的非线性相关滤波在光照变化人脸识别中的应用

针对现有的滤波算法由于光照变化而影响人脸识别性能的问题,提出了特定类子空间依赖的非线性相关滤波算法。首先,利用非线性最佳映射图像相关滤波器与非线性最佳重建图像相关滤波器之间相位的特定类子空间运算实现算法；然后,通过最小化相关平面能量、同时最大化相关波峰进一步优化；最后,利用关联分类器完成人脸识别。在扩展Yale B和PIE人脸库上的实验结果表明,本文算法在加性高斯噪声条件下仍然对光照变化不敏感,相比其他几种较好的滤波算法,本文算法取得了更高的识别率,并提高了算法执行效率。

The application of nonlinear correlation filtering based on CSSD in face recognition with illumination variation

For the issue that recognition performance of existing filtering algorithm has been impacted by illumination variation, a nonlinear correlation filtering algorithm based on Class-specific subspace dependent (CSSD) is proposed. Firstly, class specific subspace operating of phase between nonlinear optimal mapping image correlation filter and nonlinear optimal reconstruction image correlation filter is done to implement proposed algorithm. Then, it is optimized further by minimizing the surface energy and maximizing correlation peaks at the same time. Finally, associated classifier is used to finish face recognition. Experimental results on extended Yale B and PIE face database show that proposed algorithm is not sensitive to illumination variation under the condition of additive Gaussian noise. It has higher recognition accuracy and better efficiency than several advanced filtering algorithms.