PSNR, or Peak Signal-to-Noise Ratio, is the full-reference metric most engineers meet first: it counts how far a compressed video has drifted from its pristine original by measuring pixel error and reporting it on a decibel scale, where higher is better and identical files score infinity. It is computed in two steps — first the Mean Squared Error (average of the squared per-pixel differences), then a conversion to decibels as 10 times the base-10 logarithm of MAX-squared over MSE, where MAX is the largest pixel value (255 for 8-bit). Every halving of the error adds roughly 3 dB. The central catch is that PSNR is a weak perceptual predictor: it weighs an error in a face exactly like one in flat sky, models no visual masking, and gives identical scores to very different-looking distortions. It is valid only when comparing encodes of the same content and codec, which is why perceptual metrics like SSIM and VMAF were created.
The handbook · 7 volumes
Fora Soft Video Engineering Handbook
PSNR is fast and universal, but it predicts what viewers see poorly, so it works best as a sanity check next to a perceptual metric. We wrote seven volumes covering the whole video stack, and for PSNR start with Volume 1: PSNR and MSE, SSIM and VMAF, rate-quality curves and BD-rate. Written by Nikolay Sapunov, CEO at Fora Soft.
Still judging encodes by PSNR and wondering what it misses?
Send us encodes from your pipeline: we compare PSNR with perceptual metrics on your content and recommend what to measure and gate on. 250+ video products shipped since 2005.
