Artifact Filtering versus Artifact-Compensated Training in JPEG to JPEG AI Recompression
Updated
Updated · arxiv.org · Nov 10
Artifact Filtering versus Artifact-Compensated Training in JPEG to JPEG AI Recompression
1 articles · Updated · arxiv.org · Nov 10
Summary
Researchers have compared artifact filtering and artifact-compensated training methods for recompressing JPEG images into the new JPEG AI format.
While pre-filtering with neural networks reduces artifacts and file size, it increases processing time by 29 times compared to direct recompression.
Artifact-compensated training for JPEG AI achieves similar artifact reduction and compression efficiency but with much lower computational cost, making it preferable for large-scale recompression.