The AI Front Page

Reading signals from this article are folded back into your front page ranking on this device.

Research/arXiv AI/ML/August 3, 2026 at 5:33 PM

arXiv paper: Estimating SSIM from MSE for DCT-Based Compressed Images

A new arXiv AI paper by Luc Trudeau and Maria G. Martini studies Estimating SSIM from MSE for DCT-Based Compressed Images.

Research / arXiv AI/ML
Source

Follow arXiv AI/ML to make it a durable For You signal.

A new arXiv paper proposes estimating SSIM for DCT-based compressed images using only global PSNR or MSE and statistics from the reference image, rather than requiring local MSE measurements. The authors redistribute global MSE according to local variance or standard deviation to approximate local errors. Tests on the Kodak and Xiph Subset1 datasets across JPEG quality levels found that both methods produced accurate and robust SSIM estimates and substantially outperformed using global MSE alone. The approach could support more efficient quality assessment in image and compression systems, and the authors say it may extend to video by reusing reference-image statistics across multiple encodes.