OKChanging lighting
Daylight coming in through a window, an ageing light or reflections on shiny packs change the image, and the inspection reads it as a defect.
Home / Machine vision / False rejects
Guide · Quality control · Waste
False rejects in machine vision: why they happen and how to reduce them
A false reject is a good pack that the vision system rejects as bad. Each one is product thrown away or reworked, and when there are many, the operator ends up lowering the requirements or disconnecting the camera, and then the bad ones get through too. In this guide we explain where they come from and how they are reduced.
The causes
Where false rejects come from
It is almost never "the camera". It is usually one of these causes, or several at once.
The pack does not arrive the same way
Rotated packs, irregular spacing or vibration on the conveyor: the tool looks for the area where it is not.
Tolerances that are too strict
A threshold set "just in case" rejects normal variations in the pack that are not a fault.
Wrong format recipe
After a format or flavour changeover, the previous recipe stays loaded and everything is rejected.
Dirty or misaligned optics
Splashes on the lens or its protective cover, or a camera that has shifted with vibration.
The tool is not the right one
Defects that vary from one piece to the next (shape, colour, surface) are hard to describe with fixed rules; deep learning is a better fit there.
How they are reduced
Measure first, adjust afterwards
False rejects are not fixed by raising the tolerance blindly: that lets the bad ones through as well. They are fixed by knowing, reject by reject, what the camera saw and why it decided as it did.
- Store the photo and reason of every reject, and review them in a gallery
- Separate real rejects from false ones and see which tool causes them
- Adjust the threshold with the measured values, not with intuition
- Stabilise the lighting and the pack position before touching the software
- Recipes per format with user login and a log of every change
- Deep learning where the defect varies naturally
A real case
−80% waste from false rejects
On a dairy carton line, by reviewing the rejects with their photo and adjusting the inspection with that data, the waste from good packs being rejected fell by 80%, without letting bad packs through. It is not a promise for every line: each one starts from a different situation. That is why we start by measuring.
Calculate what you lose
In about two minutes, TIVOX Insight gives you an estimate of what the rejects on your line cost you each year.
We review your system
With photos of your rejects, we tell you what is causing the false ones and what can be corrected, whatever the brand of your camera.
See how it works
In the TIVOX Sentinel PLUS simulator you can see the reject gallery and the thresholds of each tool.
FAQ
Answers to the most common questions
What is a false reject in machine vision?
It is a good pack or part that the vision system classifies as defective and rejects. It is also called a false positive, because the system "tests positive" for a defect that does not exist.
What false reject rate is normal?
It depends on the product, the inspection and the requirements of each plant, so there is no figure that applies to all of them. What matters is measuring it: knowing how many of the rejects are real and how many are not, and which tool causes them.
Does raising the tolerance solve the problem?
Not blindly: false rejects go down, but bad packs start getting through. The tolerance is adjusted with the measured values of the rejects, after stabilising the lighting and the pack position.
Can you review a vision system from another brand?
Yes. With photos of the rejects and of good packs, we tell you what is causing the false rejects and what can be corrected, whatever the brand of the camera.
When is deep learning worthwhile?
When the defect or the product varies naturally from one piece to the next and is hard to describe with fixed rules: surfaces, seals, irregular shapes. For clear measurements and readings, classic tools are usually faster and easier to adjust.
Free remote review
Is your camera throwing away good packs?
Send us photos of the rejects and of good packs. We'll tell you where the false rejects come from and what can be corrected.