Tuesday, December 17, 2013

The dangers of the “Golden Template”

It’s very tempting to use subtraction from a “perfect image” to find defects. Unfortunately though, this usually results in a lot of false rejects. The reasons are many and various, but Ruben Uribe has done an excellent job of summarizing and explaining in “Change Detection for Machine Vision Applications”.

Published on the Quality Magazine website December 10th, 2013, (registration required, but definitely worthwhile,) this also suggests some steps you can take to improve image subtraction approaches. The conclusion however seems to be that none of them are much good. Instead, Mr. Uribe suggests going with “light invariant” approaches.

Anyone for 3D?

No comments: