Monday, May 30, 2011

Image processing at the speed of light

Old hands know that the fastest way to improve an image before applying blob tools or edge detection is to get the lighting right. It might be argued that the power of modern software and PC’s makes this less of an issue than it was fifteen or twenty years ago. Back then, applying a filter to a VGA image consumed significant computing resources and could add hundreds of milliseconds to the time needed to process an image, so it was essential to make the task as simple as possible.

I was reminded of this while leafing through a back issue of Photonics Spectra. “Dual Exposure Speeds Solar Cell Inspection,” (June 2010 – yes I am a very slow reader!) discusses the use of cameras from IDS (http://www.ids-imaging.com/) in a solar cell inspection application.

Two points stood out for me. The first was the use of different spectra to highlight different types of flaws. Red backlight is used to highlight through-cracks, and “diffuse incident white” lighting for surface inspection. (I assume the writer means cloudy day lighting, and yes I know white is multi-spectral.) This is interesting because I’m seeing more and more solar cell inspection done with SWIR, and red is of course getting towards that spectral region.

The second point was the comment that, “The colored light also helps simplify the image processing used to detect and measure the cracks.” Okay, you might say “Well duh!” to that, but I think it bears repeating. Simplify the image to highlight only what you want to work on and there may be no need to use computationally intensive pattern matching tools. After all, just because I have the latest and greatest electronically controlled impact driver in my tool chest it doesn’t mean I should never use a regular wrench.

Which brings me back to my opening comment. Nothing works faster than light, (even though my boss often wishes I did,) so use the fastest imaging processing tool in the known universe to simplify your application development.

No comments: