Performance verification of super-resolution image reconstruction

In this paper, the specific limitation of super-resolution image reconstruction (SRR) is reported. In many SRR studies, specially made content was used, rather than general Television and cinema content, but the amount of aliasing differs between the two. The presence of aliasing distortion is assumed during the original degradation of SRR, but a low-resolution image that we want to transform into a higher resolution may not have aliasing. It is not known whether it is possible to create a high-resolution image with SRR in the absence of aliasing, and the real power of SRR can be discussed only if general content, i.e., content that does not have aliasing, is used. The validity of SRR can be determined by examining whether new frequency components above the Nyquist frequency of the low-resolution image are created by showing a two-dimensional Fourier power spectrum of the output. This has not been performed in previous studies. In this study, we perform frequency analysis for SRR assuming that general content does not have aliasing. From the results of our experiments, in conjunction with the frequency component, we prove that SRR does not work with general content.