Akademik Veri Yönetim Sistemi
SIGNAL IMAGE AND VIDEO PROCESSING, cilt.16, sa.3, ss.661-668, 2022 (SCI-Expanded)
This research proposes a novel approach to improve the performance of the very deep super-resolution (VDSR) method in single image super-resolution (SISR). In the literature, VDSR is one of the most popular CNN-based super-resolution approaches and this study aims to improve the performance of the VDSR network structure's output by using the natural image quality evaluator (NIQE). New coefficients have been proposed to improve performance when combining the residual image obtained from the VDSR method and the original low-resolution image. The NIQE parameter is defined as an optimization parameter with these proposed coefficients, and it is aimed to be minimized in each SISR application. The particle swarm optimization method was used to determine the optimal coefficients for improving SISR output in this work. In the last stage, the HR image was obtained by equalizing the histograms of the low-resolution image with the image output that was optimized according to the NIQE parameter. The outputs of the proposed approach were compared with other SISR methods in the literature. It has been observed that effective use of NIQE assessment, which is one of the no-reference image quality assessment (NR-IQA) methods, increases the output image quality in SR application.