Plenty of tools in the box

How many tools does a machine vision software package need? According to “Applying Algorithms,” published in the November ’08 edition of “Vision Systems Design,” not very many.

The thrust of the article, written by three well-informed people from Soliton, is that most vision problems can be solved with just four fundamental tools. No, I’m not going to tell you what they are. You’ll have to read the article for that.

Of course, there may be a “chicken-and-egg” element to this analysis: perhaps the problems are selected to suit the tools available rather the tools being picked to suit the need.

The authors go on to note that many engineers working on a vision application tend to reach straight for the pattern-matching tool, treating as a something of a panacea. However, one major disadvantage of this approach is that the algorithm is essentially a black box. In other words, if it gives the occasional false result it’s very difficult to diagnose the reason why.

This is something I agree with. Pattern-matching is very powerful but it removes the user from the actual image processing. Sticking with the basics, like edge detection, for example, allows the developer to see why a particular image resulted in a false positive. But with pattern-matching the cause is hidden.

On the other hand, I’m always advocating that the software developers should focus on ease-of-use, so perhaps what I’m really suggesting is that pattern-matching needs greater transparency. Being able to observe the inner workings might make debugging easier.

Something for you experts at Cognex, Matrox, NI, etc. to consider.