Fields of Technology for Machine Vision – Part 2

In my last post, I discussed technology names related to machine vision, to help you navigate this business. To recap, this Venn diagram illustrates computer vision’s focus on general vision algorithms and machine vision’s practical application of these and other algorithms to vision tasks that are too fast, precise, or tiresome for human vision. The ovals inside Machine Vision represent some of the technologies used in machine vision. Today, I’ll talk about image processing, both optical and digital.


Image processing improves an input image for use in subsequent processing. In machine vision, image processing is used to amplify or select image components, such as object edges or image areas with specific intensities. Teledyne DALSA’s Sherlock™ machine vision software uses the term preprocessor for image processing functions, as they are applied to an image before extracting information from it.

Lighting and optics in a machine vision system also perform image processing. For example, low-angle illumination – putting the light nearly parallel to the surface of an object – highlights surface flaws. The low-angle illumination “catches” on things like dust and scratches, making them bright against the dark background where the object surface is flat.

The left image shows a scratch on a cell phone case taken using low-angle illumination to amplify the scratch signal. This optical image processing also amplifies the natural texture of the case.  The middle image is the result of applying a Teledyne DALSA Sherlock preprocessor that amplifies local intensity changes that are often caused by object edges. This digital image processing makes both the scratch and the background texture more obvious. The right image shows the result of additional image processing to remove the bright spots due to the case’s texture while maintaining the scratch signal. The point here is that lighting and optics are important image processors – if the vision system could not “see” this scratch, the digital image processing would fail.

Image processing has a long history of improving images for human viewing and, of course, for producing fake or trick images – see the supermarket tabloids. For example, a photographic darkroom technique known as unsharp masking subtracts a blurred version of an image from the image to enhance high spatial frequency components and so make the resulting image look “sharper”.

Starting in the 1960’s digital image processing methods got a major boost from the U.S. space program.  Old books by Castleman and Gonzalez tell how images from Ranger moon probes were processed to make them useful for scientists and, I suppose, to encourage support for the space program.  Facebook’s recent purchase of Instagram for one billion US dollars shows that image processing has incredible value (or at least pre-IPO Facebook stock value). There’s a good chance that some of Instagram’s technology has space program ancestry, so it would be a fitting turn-about if NASA used Instagram filters on images from Curiosity on Mars… just to encourage further support for the Mars rover program.

’til next time,



About Ben

I earned M.S.E.E. and Ph.D. degrees from Stanford, was at MIT for many years, and have been working in the vision business longer than I want to admit. At Teledyne DALSA, I develop vision algorithms, provide customer support on difficult vision tasks, and do "technical marketing" -- writing papers, blogging, and lecturing.
Posted on by Ben. This entry was posted in Image processing, Machine Vision, Software and tagged , , , . Bookmark the permalink.

Comments are closed.