There are many advantages to being a journalist reporting on the latest developments in machine vision and image processing. Although perhaps the size of the paycheck falls at the bottom of this list, being involved with people in the industry that are very intelligent must rank close to the top. Not only are such people invaluable resources when it comes to preparing technical articles, the knowledge they impart can be especially useful in evaluating newly introduced products and systems.
Machine vision requires an understanding of optics, lenses, sensors, cameras, lighting, computers, algorithms, and mechanical design. Obtaining this knowledge base can be especially intimidating for newcomers to the industry. Unlike the technical press, the popular press does little to promote an understanding of the subject and can sometimes be a hindrance rather than a help for those wishing to understand the technicalities of such systems.
Just last month, Nokia unveiled its 808 PureView smart phone that features a 41-Mpixel imager. "Roll that over on your tongue for a minute," wrote Ned Potter at ABC News. "Forty-one megapixels. Forty-one. That's not a camera, that's a statement." Others, of course, were quick to add their megapixels' worth. "It's a phone that has so many megapixels, its megapixels have megapixels," wrote Lynn La, CNET's associate editor for cell phone and smart phone reviews.
Those new to the industry may wonder why vendors in the machine-vision market can charge $5000 and more for cameras with fewer pixels while the Nokia 808 costs a mere $600. Given just these facts, it does seem rather unfair, but a closer look may demystify the subject.
If my knowledge of mathematics serves me well, the Nokia camera's 7728 × 5368-pixel, 1/1.2-in. imager corresponds to a 1.4-µm pixel size. Coupled with a custom f/2.4 lens from Carl Zeiss, the smallest point to which a lens can focus a beam of light (at 580-nm yellow light)—the Airy disk—will be 3.4 µm. Obviously, this is a lot larger than the 1.4-µm pixel size featured in the Nokia camera's sensor!
Thus, although it may at first appear that reducing the pixel size of such imagers may result in increased resolution, the diffraction-limited nature of light itself proves this to be a falsehood. To form an image from this imager, Nokia oversamples the pixels, combining nearest-neighbor pixels that result in a final image being approximately 5 Mpixels in size.
Needless to say, I have not seen any reviews that specify the modulation transfer function of the camera system, which is the specification you need to make any kind of comparative judgment between Nokia's camera and the more expensive cameras intended for machine-vision applications. All that Nokia has produced are images of some colorful climbers taken on a very bright, sunny day. For the company, this was also another really good marketing idea.
Those well-established in the machine-vision and image-processing community are aware of such marketing practices. Well, if you are new to the industry and are asked why your machine-vision or image-processing camera costs a lot of money, you could always say that bigger is better.
|Andy Wilson, Editor in Chief|
Vision Systems Articles Archives