Fig. 1. classification of super-resolution reconstruction method

Fig. 2. Feature extraction results of different convolution kernels: (a) original， (b) Laplacian， (c) Laplacian + horizontal gradient， (d) Laplacian + vertical gradient， (e) Laplacian + horizontal second gradient， （f） Laplacian + vertical second gradient， (g) horizontal gradient， (h) vertical gradient， (i) horizontal second gradient， (j) vertical second gradient.

Fig. 3. Dictionary training process

Fig. 4. Comparison of sparse dictionary (a) before; (b) after

Fig. 5. Comparison of image noise: (a) noise of LR image， (b) noise of reconstruction image by 4 times

Fig. 6. Saliency algorithm based on sparse features

Fig. 7. The process of saliency regional selective super-resolution reconstruction algorithm

Fig. 8. Training dataset

Fig. 9. Comparison of reconstruction by different algorithms (a) LR image, (b) bicubic interpolation, (c)ScSR, and (d) proposed algorithm

Fig. 10. Results of super-resolution reconstruction (a)Yang algorithm, (b)SRCNN, (c) saliency-super-resolution, (d) saliency map

Fig. 11. Contrast of gray gradient value of adjacent pixels　note: the x-coordinate is the x-coordinate value of the image pixels

Fig. 12. Comparison of background noise suppression results of different methods (a) LR noisy image, (b) ScSR, (c)SRCNN, (d) bicubic interpolation(BI), (e) median filtering(MF), (f) gaussian filtering(GF), (g) bilateral filtering(BF), (h) proposed algorithm(PA)

Fig. 13. SNR comparison of background noise suppression results of different methods.　Note: the ordinate is the value of SNR