A UHD TV, for example, is the state-of-the-art digital television
in the consumer market. It adopts various image restoration
algorithms to enhance the quality of low-resolution image contents. A low-end UHD TV uses a finite impulse response (FIR) type
image restoration filter for cheap, easy implementation, while a
high-end UHD TV with sufficient amount of memory may adopt
an example-based image restoration algorithm. In practice, the
truncated constrained least squares (TCLS) filter is suitable for
enhancing the UHD TV images because of the finite filtering support for real-time restoration [5,6]. However, the TCLS restoration
filter is sensitive to the accuracy of estimating the point spread
function (PSF). Since a perfectly accurate estimation of the PSF is
almost impossible, the TCLS restoration filter cannot successfully
restore the high-quality image in the commercialization level by
itself. Alternatively, example-based algorithms are widely used in
image restoration and super-resolution (SR) [7–9]. Although
example-based image restoration can provide well-restored
images by replacing the low-resolution patch with the appropriately selected high-resolution patch, its disadvantage is twofold:
(i) generation of the patch dictionary and search of the best patch
require high computational load, and (ii) patch mismatch error
produces undesired artifacts in the restored image.