SSIM
SSIM (structural similarity index) is a video and image quality metric that compares a compressed frame with its original by looking at brightness, contrast and structure, rather than raw pixel differences. Scores usually top out at 1, which means identical. SSIM tends to match human judgment better than PSNR, because people notice structural changes like blur and blockiness more.
Why it matters
SSIM gives encoding teams a quick quality check that tracks human judgment better than raw pixel error. It helps catch blur and blocking that hurt how premium content looks. Because it is cheap to run, it suits automated tests across large libraries.
How it works
- Small windows of the compressed frame are compared with matching windows in the original.
- Each comparison looks at brightness, contrast and structure, meaning the pattern of edges and textures.
- The three comparisons are combined into a similarity value for each window.
- Window values are averaged into a frame score, then averaged across frames for the video.
What is a good SSIM score?
A good SSIM score is close to 1, which means the compressed frame is nearly identical to the original. Values around 0.95 and above usually look very close to the source, while lower values point to visible blur or blocking. The scale is squeezed near the top, so small differences such as 0.97 versus 0.99 can still be noticeable.
SSIM vs PSNR: what is the difference?
PSNR measures how far each pixel has moved from the original, while SSIM measures whether structure, such as edges and textures, has been kept. SSIM usually matches human judgment better, because people notice blur and lost detail more than tiny uniform shifts in brightness. PSNR is simpler and faster, so many teams report both.
Where it's used
OTT services and live sports streamers use SSIM in encoding tests to confirm that compression keeps edges, textures and fine detail intact, often reporting it beside PSNR and VMAF.
By the numbers
- SSIM was published in April 2004 in IEEE Transactions on Image Processing by Zhou Wang, Alan Bovik, Hamid Sheikh and Eero Simoncelli. (Wikipedia: Structural similarity index measure)
- The authors of the original SSIM paper each received a Primetime Engineering Emmy Award in 2015 for its wide use in the television industry. (Wikipedia: Structural similarity index measure)
