Bilibili Tech
Feb 24, 2023 · Artificial Intelligence
Understanding Video Super-Resolution: Principles, Common Defects, and Practical Enhancement Techniques
Video super‑resolution, pioneered by deep‑learning models such as SRCNN, can synthesize plausible high‑frequency detail but often introduces artifacts like loss of stylistic noise, inconsistent line depth, texture smearing, and temporal flicker, which can be mitigated through preprocessing (BM3D denoising, descaling), targeted post‑processing (Gaussian blur, unsharp masking) and selective edge‑based texture merging to preserve original artistic style while enhancing perceived sharpness.
BM3DCUGANFourier transform
0 likes · 13 min read