How different are these two, really?
"Can you tell?" is usually answered by squinting. This answers it with a number.
"Can you tell the difference?" is the question behind every compression decision, and it is usually answered by squinting at two images in separate tabs. This answers it with a score and, more usefully, a map: the difference between the two is amplified until it can be seen, which shows where an encoder actually spent its errors.
What you get
- A similarity score, using SSIM rather than raw pixel difference.
- A difference map, multiplied so faint changes become visible.
- The size of each file and how much smaller one is.
- The largest single difference anywhere in the picture.
Reading the map
- Black means the two are identical there.
- Bright areas are where they diverge most.
- Edges and fine texture light up first, because that is what encoders discard.
- Flat areas stay dark even at low quality, which is why they fool the eye.
Questions
How do I tell if a compressed image lost quality?
Compare it against the original and measure rather than guess. SSIM tracks what people actually notice: above about 0.99 the difference is invisible outside a pixel-by-pixel comparison, and by 0.95 you can find it if you know where to look.
Why do the two images need the same dimensions?
Because overlaying different shapes compares the geometry rather than the encoding. If one has been resized or cropped, the score would be measuring your own decision, so it declines to give one instead of reporting a meaningless number.
Why is the difference map so bright?
It is amplified about six times. The real differences between a good compression and its original are a handful of values per channel, which is invisible on screen, so the map shows the shape of the error rather than its true intensity.