If you're considering color machine vision software...
If your inspection task involves only recognition of uniformly colored objects in a uniform lighting environment you probably don't need to read this page. In fact you might not even need color machine vision.
Otherwise, before risking your career on an expensive vision system you might want to do some tests.
Can the system differentiate multicolored, randomly oriented and/or complex-shaped objects?
Multicolored patterns are characteristic of many fabrics, natural materials, pharmaceutical capsules, and manufactured components. The nine simple two-color patterns to the left are designed to test a system's ability to differentiate such materials. The vision system should be able to reliably recognize each pattern based on the entire pattern, and/or only a small part. Some systems using traditional color vision approaches have great difficulty with this test.
The pattern identification numbers are not a mistake. After all, human inspectors are not troubled by labels, printing etc. You should expect no less of your color machine vision system.
It takes less than 5 minutes to manually train WAY-2C to recognize all patterns in the above test image. It takes only a few seconds for the same training from a prepared script file.
Is new training required on more than one class every time a class is added, substituted, or removed?
WAY-2C usually requires training only on the new class, or when the camera, lighting, or framegrabber is changed.
Downloading images for testing.
You are authorized to download any images on this website for the sole purpose of vision system evaluation. Publication in any form other than short excerpts for review purposes requires the express written permission of the author.
All our images have been obtained from actual inspections or inspection demonstrations. All images (except for the specially constructed test patterns) will have lost a little information as a result of compression (to JPG or GIF) but still should be adequate for most tests.
We also have a secret cache of unpublished test images available on request in case you want to verify that a potential system supplier has not made up special "ad hoc" solutions for our published examples.
From your browser click on the selected format.
For BMP, JPG or GIF formats a full sized version of the image
will appear on on your screen.
Right click on the file and select "Save Image As..."
For other file formats your browser will prompt you regarding saving the file.
Once the image is saved, if necessary you can use an image file conversion utility to convert it to the machine vision system's preferred file format for testing.
If your vision system will not accept images from files:
Either download the image file as described above and print it off-line
or connect your system to a good color printer and select "File: Print".
Then place the image in front of your vision system's camera for testing.
Note: Because of the loss of some information and the possible introduction of irrelevant color information in the printing process,
testing printed images will not be as reliable as direct testing of the image file..
Color based recognition done correctly can be easy, robust and reliable. Color based recognition done incorrectly can be tedious, frustrating and often unreliable.