ERQA: Edge-Restoration Quality Assessment for Video Super-Resolution

Abstract

Despite the growing popularity of video super-resolution (VSR), there is still no good way to assess the quality of the restored details in upscaled frames. Some SR methods may produce the wrong digit or an entirely different face. Whether a method's results are trustworthy depends on how well it restores truthful details. Image super-resolution can use natural distributions to produce a high-resolution image that is only somewhat similar to the real one. VSR enables exploration of additional information in neighboring frames to restore details from the original scene. The ERQA metric, which we propose in this paper, aims to estimate a model's ability to restore real details using VSR. On the assumption that edges are significant for detail and character recognition, we chose edge fidelity as the foundation for this metric. Experimental validation of our work is based on the MSU Video Super-Resolution Benchmark, which includes the most difficult patterns for detail restoration and verifies the fidelity of details from the original frame. Code for the proposed metric is publicly available at https://github.com/msu-video-group/ERQA. a https://orcid.

Hybrid Refractive-Diffractive Lens with Reduced Chromatic and Geometric Aberrations and Learned Image Reconstruction

“…The FEL criteria are based on edge detection in real-world images and do not require a ground truth image. In work [ 37 ], a method for evaluating the quality of detail restoration in video super-resolution was proposed, called edge restoration quality assessment (ERQA), which was also based on edge detection. While this work showed that edge restoration is critical for human perception of detail restoration, their method required ground truth images.…”

Compressed Video Quality Assessment for Super-Resolution: A Benchmark and a Quality Metric

“…The ERQA metric [13] assesses images in terms of detail restoration. To do so, it detaches object edges and matches them with their counterparts in the reference image.…”

“…After super-resolving videos, we calculated the following objective-video-quality metrics on the results: PSNR, MS-SSIM [26], VMAF [2], LPIPS [32], and ERQA [13]. We ranked the SR methods by calculating BSQ-rate (bitrate-forthe-same-quality rate) [33] for each SR+codec pair relative to base codec performance, where the base codec is the one we used to compress low-resolution video.…”

“…The next step was to apply each SR method to each video, thereby producing the dataset of distorted videos. We also compressed [2] 0.344 0.448 LPIPS [32] 0.414 0.431 ERQA [13] 0.582 0.624 MDTVSFA [15] 0.634 0.644 ERQAxMDTVSFA 0.770 0.801 Proposed 0.821 0.838 the source videos without downscaling and added the results to our dataset. We conducted a subjective comparison like the one in Sec.…”