超分辨率人脸图像重构识别

Reconstruction and recognition of super-resolution face images

为了解决单幅低分辨率人脸图像重构问题,提出了基于线性物体类理论重构超分辨率人脸图像的新方法。首先利用ICA和PCA提取不同分辨率人脸的特征子空间,然后利用通过训练得到的分辨率转换矩阵重构其相对应的超分辨率人脸图像,实验表明该算法与传统的算法相比重构出的人脸图像质量和识别率都有了很大的提高。

A new super-resolution face image reconstruction method based on linear object-class theory is proposed to deal with the problem of low-resolution face images.First,the feature subspaces are formed from different resolution images using ICA and PCA.Then the low-resolution image is transformed into its corresponding super-resolution face image using the transformation matrix predetermined by learning.Experimental tests showed that both the reconstruction quality and the recognition rate are improved greatly with respect to normal algorithm.