如何提高人脸图像的分辨率是计算机视觉中的经典问题。在视频监控中,由于目标人物距离摄像头较远,得到的往往是低分辨率的人脸图像。针对此问题,提出一种结合主成分分析(PCA)和最大后验概率(MAP)的人脸超分辨复原算法。利用PCA模型获得高分辨率人脸库的特征;通过MAP计算输入的低分辨率人脸图像在这些特征上的表示系数,并重建出其对应的高分辨率特征;对重建的特征进行约束增强;结合高分辨率人脸库的平均向量得到最终超分辨率复原的图像。为了验证本文算法的有效性,将AR人脸库中图片分别用本文算法与其他算法放大4倍,结果显示无论是在视觉效果还是在评价指标方面,本文算法都优于其他方法。本文算法不仅提高了人脸图像的分辨率,还更好地保持了图像的边缘信息。

How to improve the resolution of face images is a classic problem in computer vision. During video surveillance, since the target person is faraway from the camera, the result is often a low-resolution face image. Aiming at this problem, we propose a face super-resolution restoration algorithm combining principal component analysis (PCA) and maximum a posteriori probability (MAP). Firstly, we get the characteristics of the high-resolution face database based on the PCA model. Secondly, we calculate the representation coefficients of the input low-resolution face images on these features by MAP and reconstruct the corresponding high-resolution features. Thirdly, we make the constraint enhancement of the reconstructed features. Finally, we obtain the final super-resolution restoration images based on the average vector of high-resolution face database. In order to verify the effectiveness of this algorithm, we make the experiments that the images in the AR face database are amplified four times using this method and other methods. The result of this method is superior to other methods in either visual effects or evaluation indicators. This algorithm not only improves the resolution of face images, but also maintains the edge information of the image better.