Frequency domain analysis of Super Resolution Image Reconstruction and super resolution with nonlinear processing
Seiichi Gohshi · 2016
Super Resolution (SR) is an interesting topic in image and video research. Among SR Super Resolution Image Reconstruction (SRR) is one of the most common SR technologies. Originally SRR was proposed for still images. Recently SRR has been applied to video. However, there are important differences between still images and video that must be addressed when working with SRR. The basic hypothesis of SRR is that, using several low-resolution images, we should be able to construct a single high-resolution image. However, adjacent video frames cannot always guarantee this result. These limitations and image quality for video have been extensively examined in the time domain as subjective assessments. In this study, analysis of SRR is conducted in the two dimensional frequency domain. The limitations of SRR is discussed with two-dimensional fast Fourier (2D-FFT) results as objective criteria. To overcome the limitations of SRR Nonlinear Signal Processing (NLSP) is proposed in this paper. Although it is a simple algorithm, it can create higher frequency elements that the input image does not have. Real time hardware with NLSP is also mentioned.