PROJECT TITLE :

Noise-Robust Iterative Back-Projection

ABSTRACT:

As a result of denoising, noisy image super-resolution (SR) is a substantial challenge. There is no clean reference image for iterative back-projection (IBP), which can help improve the reconstructed SR image further. Back-projection algorithms for noisy picture SR are proposed in this work. Its primary purpose is to ensure that LR and SR photographs are identical. With the use of noisy and denoised reconstruction errors, we are attempting to estimate the clean reconstruction error. To estimate the clean reconstruction error, we devise a novel cost function on the PCA transform domain. Texture probability is used to blend noisy and denoised reconstruction mistakes in the data component of the cost function. As a result, the Laplacian properties of the reconstruction mistake are integrated into the regularisation term. Eigenvector estimation is also proposed to reduce the impact of noise. Using the proposed method, back-projection can be done more reliably than with the usual IBP, and it may be used as a post-processing method for any other SR technique.