How we decide "looks the same"
Updated
PicPinch doesn't guess at quality. Every compressed file is decoded again and scored against the original with SSIMULACRA2, a measure built to match what people see, both on the whole image and on every small part of it. A file is kept only if the score reaches the level you chose: 80 by default, 85 for High or 90 for Highest. If no smaller file reaches it, you get your original back.
Why not just "quality 80"?
The quality slider in an image editor is an instruction to the encoder, not a measure of the result. The same setting gives very different results on different images: quality 75 can be invisible on a busy photo and plainly blocky on a screenshot with text. And JPEG's quality numbers mean something different from WebP's. So instead of choosing a setting and hoping, we measure the result and adjust until it reaches the target.
The score, in plain words
SSIMULACRA2 compares a compressed image with its original and returns a number, at most 100 for an identical image. Its authors describe what the levels mean in its README, for an average person at a normal viewing distance, looking at the image at its real size:
| Score | What it means | In PicPinch |
|---|---|---|
| 70 | High quality: differences are barely noticeable side by side, and not annoying | Never this low |
| 80 | Very high quality: no difference noticed when the two are side by side | Standard (default) |
| 85 | Excellent: no difference noticed even when one replaces the other in place | High |
| 90 | Visually lossless: no difference noticed even when they flicker back and forth | Highest |
Differences can still show if you zoom in a lot, which is what High and Highest are for.
The worst part, not the average
A score for the whole image is an average, and an average can hide a problem: a clean sky can make up for a smeared face. So we also cut every image into squares of about 256 pixels and score each one. A file's score is the lower of two numbers: the whole-image score, and the worst square's score plus a 5-point allowance.
Before we added this check, 24 of our 204 test images (12%) had some area below 75 at the default target, the worst at 69: mostly clean images with one badly compressed patch. With the check, none do, and the files grew by only about 1.5% with PNG and JPEG. In our current benchmark, the worst square in any file scored 75.1 at the default, 80.2 at High and 86.8 at Highest.
Never bigger, never worse
- Every file is decoded again and scored before you see it. None is below its target.
- If nothing smaller reaches the target, you get your original back, untouched.
- If a result would be bigger than your original, you get your original back, unless you asked for a particular format, such as on a converter page.
See it for yourself
Open Compare on any result and use the slider, or hold the button to see the original. Then press "Most-changed spot": it zooms straight to the square that changed most, so you can judge the worst case with your own eyes instead of trusting a number.
An honest note
The scorer we report with is the same one our engine aims for, which favours it slightly in comparisons. That's why our benchmark describes how we measured, lists its limits, and why Compare lets you check any file yourself.
Questions
- Which quality should I choose?
- The default (80) for websites, chats and forms. High (85) if people will zoom in. Highest (90) for prints or images you'll edit further.
- Is 80 out of 100 "80% quality"?
- No. It's a score on SSIMULACRA2's scale, where 80 already means no difference noticed side by side. It isn't a percentage, and it isn't the same as an editor's quality 80.
- Why did I get my original back?
- Because no smaller file reached your quality level. It's common for images that were already well compressed, and it means your file is already about as small as it can get at that quality.
More detail: how it works, the benchmark.