Name
Title | ||
|---|---|---|
Single-image super-resolution based on regularization with stationary gradient fidelity |
Abstract | ||
|---|---|---|
Single-image super-resolution, reconstructing a high resolution (HR) image from a low resolution (LR) one, is an ill-posed problem. Regularization, a method for ill-posed problem, is widely used in super-resolution. The definition of the fidelity term in regularization energy is a key to the performance of regularization based super-resolution. Traditional fidelity is defined as the energy of error image between down-sampled version of HR image estimated and the observed LR image. Since the mpixels not in sampled positions are not considered, the gradient of traditional fidelity is non-stationary. To avoid this problem in traditional fidelity, we propose to define the fidelity term as the energy of error image between the estimated HR image estimate and its blurred version. When gradient decent is used to optimize the regularization energy, we propose an error interpolation fidelity gradient (EIFG) method to estimate a stationary gradient of proposed fidelity. Compared with other methods, experimental results show that our method improve both qualitative and quantitative performances of reconstructed HR image. |
Year | DOI | Venue |
|---|---|---|
2017 | 10.1109/CISP-BMEI.2017.8301942 | 2017 10th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI) |
Keywords | Field | DocType |
super-resolution,ill-posed problem,regularization,fidelity,stationary gradient | Iterative reconstruction,Computer vision,Fidelity,Gradient descent,Computer science,Interpolation,Regularization (mathematics),Artificial intelligence,Image resolution,Superresolution | Conference |
ISBN | Citations | PageRank |
978-1-5386-1938-4 | 0 | 0.34 |
References | Authors | |
0 | 5 | |