Fig. 1. Face image decomposition

Fig. 2. Algorithm framework (a) fuzziness and down sampling; (b) face decomposition based on PCA; (c) MAP reasoning to get the best feature face; (d) constraint enhancement

Fig. 3. Test images

Fig. 4. Image processing results by different algorithms (×4). (a) Input images; (b) Bicubic method; (c) EigTran method; (d) ScSR method; (e) SRCNN method; (f) proposed algorithm; (g) original images

Fig. 5. Contrast experiment at n=819. (a) Input images; (b) Bicubic method; (c) proposed algorithm; (d) original image

Fig. 6. Effects of the number of training samples n on super-resolution restoration image PSNR

Fig. 7. Effects of the number of training samples n on super-resolution restoration image MSSIM

Fig. 8. Low-resolution face image processing without glasses. (a) Original high-resolution images; (b) low-resolution input images; (c) train set-1; (d) train set-2

Fig. 9. Low-resolution face image processing with glasses. (a) Original high-resolution images; (b) low-resolution input images; (c) train set-1; (d) train set-2