The goal in this lecture was to learn, how Images are processed and computed in our brain. Furthermore the understanding for image processing in the computer should be better and some fundermental processes (Gabor Filters, Edge- and Corner Detectors, Saliency Detectors…) were discussed. In the exercises we should take any pictures and use Matlab or open-cv to do something “interesting”.

First me and my group struggeled a bit with the specification, but then we found some really interesting. We decided to take photos from Printed Circuit Boards and detect several Surface Mounted parts on them. And this was quite hard. Seperating the parts from the Board with all that conducted paths on it, or different sizes (e.g. for SMD Resistors). We decided to detect 0603 and 0805 case style Elements, and classify detected elements as Resistors or Capacitors.

This was pretty hard, because you have to imagine, that if you are only looking onto a black and white image, some 0603 resistors can not be differenced from pads of unequipped parts. After hours of hard work the results can be shown:

For that shown whole PCB we got very good results:

correct positives: 109 (out of 123)
false positives:     18
false negatives:    14
positive detection rate: 88,62%
precision:                    85,83%

Here a short comparison of the classification preformance to other Boards, where the results are not as good, but its still very impressive how good that works: