Beyond monochrome machine vision Karratha WA

In the world at large, monochrome is largely passé. Even the cheapest TVs today feature color, and have for many years. A variety of military applications are still in monochrome, but that is typically a function of the limited wavelengths used in night vision applications. Monochrome is still common in the CCTV universe, but a shift is under way there toward both color and higher resolutions. Then, there is machine vision. For some of the most advanced imaging applications in common use, machine vision is still, largely, a monochrome world—and with good reason.

Imaging in industrial production is typically used for such tasks as quality control. The general goal is to identify a feature or features on an object, such as a circuit board, to determine if something is missing or has been improperly aligned during the manufacturing process. Common machine vision requirements like edge detection and blob (lighter vs. darker) analysis often do not require color, and can benefit in many cases from monochrome. These processes typically work to maximize the data relevant to the specific task at hand, while minimizing unnecessary background data. Color generally generates a higher data volume, requires more data processing, has reduced sensitivity and a somewhat reduce spatial resolution which can work against specialized machine vision functions. There are a variety of additional color issues that come in to play, including white balance and lighting, that can affect the performance of a color camera but have less impact on the monochrome camera.

Monochrome can even handle various color-related machine vision tasks more effectively than a color camera. "Quite often, if somebody is just looking for a 'green mark' or a 'red mark,' rather than using a color camera, this can better be handled using a monochrome camera with single color filters or a combination of lighting to better discriminate the color of interest from the background," said Marty Furse, CEO, Prosilica Inc. (Burnaby, Can.). "If you need to align a green stripe with a triangle shape, you can sometimes achieve that better by enhancing the green with a green light and using a filter to subtract the background."

From a bandwidth and processing standpoint, the differences between color and monochrome can be a toss up. "When you have Bayer output, which is not interpolated data, you have the same amount of data as monochrome so it doesn't really impact the bandwidth, but in this case, the host computer usually needs to perform color interpolation before image processing begins," said Furse. "There are the various color interpolation modes: YUV 411, YUV 422, YUV 444 and RGB 24, which is an interpolated 24-bit color. Bandwidth can become an issue when using interpolated modes. Generally speaking, RGB 24 pushes at least three times more data than monochrome, but the host computer does not need to interpolate the resulting data prior to processing. Thus, there is a trade-off between interpolated and un-interpolated modes because you're trading off host computer loading versus interface load. It is not always a clear decision as to which is better. It tends to be application-specific and interface specific choice." Furse added that with GigE, the bandwidth is high enough that to run the camera at full speeds even with interpolated color.

With the cameras themselves, there is the option to go with a one-chip or three-chip solution. A one-chip camera is typically much simpler in design, featuring a single sensor element and a Bayer pattern filter that provides 50 percent green, 25 percent and 25 percent blue. The green value is doubled to reflect how the human eye resolves green to provide for a similar result. Bayer output does have some negative issues, such as a loss of data. The answer is to use interpolated color which relies on algorithms to fill in the missing data, but a cost in bandwidth and processing power. Both monochrome and color single-chip systems are roughly comparable in price, and in fact a single-chip color system can be cheaper give the high-volume production of this format in consumer markets.

A three-chip camera uses three sensors, one each for red, green and blue. The incoming light is focused on each sensor by a beam-splitting prism, with each separate beam then passing through the appropriate red, green or blue filter. This added complexity comes at a cost, with the added expense from the additional sensors, the prism system and the specialized optics.

The choice between three-chip and one-chip solutions will largely be driven by the application. There is a general move from three-chip to one-chip systems, fostered by the increases in resolution brought on by shrinking pixels and increasing densities in combination with a notably lower price. "The bulk of the color demands are for single sensor in machine vision, mostly to do with cost," said Furse. "We don't receive too many requests for three chip color anymore. They used to be a lot more common, but I think concerns about resolution have largely disappeared."

However, a variety of applications absolutely require the greater color accuracy. "One of our bigger customers that needs three-chip technology is inspecting CDs, which are checked for true color fidelity within the print of the CD label," said Gary Pitre, eastern regional sales manager, Toshiba America Information Systems Inc. (Irvine, Calif.), Imaging Systems Division. "We also have some customers in the semiconductor industry where color happens to be important to them. Outside of machine vision, we've had some decent success in the broadcast market with reality TV for things like helmet cams. The World Poker Network uses those three-chip color cameras that are hidden inside the poker table to view each player's hand because they want to be as close in quality as they can be to their standard production cameras."

COLOR MACHINE VISION APPLICATIONS

It is not uncommon for people looking for a machine vision solution to assume that it should automatically be color. "Probably half of the color requests that we receive are due to a vague preference for color —a feeling that somehow color is better or more flexible," said Furse. "It sometimes takes a brief explanation about what is lost by choosing color over monochrome."

But even with monochrome's general advantages, color imaging can play a role in more traditional machine vision applications where color is being used as an indicator and the color hue is subtle. "For the pure recognition of parts, and measurements and blob analysis and so forth for machine vision there is not always an advantage to use color, because you have the full information and black-and-white," said Dr. Joachim Linkemann, product manager, vision components, Basler AG (Ahrensburg, Ger.). "However, if you have a slight red tone on the surface and a slight green tone, in monochrome they might look very similar whereas red and green represent a huge difference in color."

From an application standpoint, there are many processes where shape has no real importance, but color is critical. Agriculture and food production is an obvious area where this comes into play. "Color is important in the food industry for checking fruits, vegetables and grains where discoloration often indicates product defect," said Linkemann. "The color information is absolutely necessary and in some cases you work with near infrared, such as checking for over-ripe bananas. It can also be important with inspecting wood laminates."

Color is critical, in a life and death manner, with some pharmaceutical applications. There are a variety of methods to discover if a tablet was left out of a pack, however, how do you detect if the wrong tablet (an entirely different drug) of a nearly identical shape was mistakenly inserted? "If you're doing a tablet filling operation, many of these tablets are very similar in size and shape and real close and color in terms of intensity," said Pitre. "But the hue may be a little bit different, and color imaging can be important where the hues are subtle. It's very critical to make sure that there are no foreign tablets that could end up in the wrong package." Pitre added that color can similarly help identify the contents of packaged items like medical emergency kits that have a variety of similarly shaped but different colored objects.

As noted earlier, printing operations also naturally benefit from color imaging, and color is also critical in applications like recycling where the goal is to sort glass or plastic items by color as they run down a conveyor line with random shapes and sizes.