Convolutional Neural Network Combined with Half-Quadratic Splitting Method for Image Restoration
Algorithm 2. Single image super-resolution.
1. Read the high-resolution image (HR), cut the size to integral multiples of the scaling factor of , so that the down-sampling is changed to low-resolution image (LR);
2. Change the LR of RGB space to YCbCr space to get the image LR_ycbcr, and take the LR_y of component Y as the value in Eqn. (14);
3. Restore the HR_bic of the original image size with Bicubic interpolation for LR
4. Change the HR_bic to YCbCr space to get the image HR_bic_ycbcr, and take the component Y as the iterative output, i.e., initialization of in Eqn. (14);
5. for i =1:m (m is the number of iterations)
6. for j =1:n
7. Execute the Eqn. (14) and continue to update the to
8. end
9. Load relevant level of denoiser according to iterative value i and take the obtained as the model input
10. The model output is residual and get the final output after making difference
11. end
12. Take the output as the component Y of HR_bic_ycbcr and then convert it to RGB space to get the final reconstruction result
