- » View All
Black and White Successes Lead to Colorful Opportunities
by Winn Hardin, Contributing Editor - AIA Posted 01/20/2004
Although the 'real world' may be in color, machine vision has steadfastly focused on a black and white (BW) world. Color typically is viewed as a distraction in machine vision – something to be eliminated with lighting and filters. Color vision systems use more expensive color cameras that produce three times the data of their monochrome counterparts, adding to the systems data processing requirements (and cost) and the complexity of the lighting solution (and cost). So who needs an expensive distraction?
Success often brings its own challenges, however, and the vision industry is no different. Manufacturing managers, enjoying the labor cost savings from a working machine vision system on their plant floor, are pushing for automated solutions that solve their color inspection conundrums. Vendors are responding with enhanced color systems that use special calibration, optics, algorithms and other methods to manage the exponential complexity of the color vision application.
Like any machine vision system, the success of a color vision system depends heavily on lighting techniques. The color of an object is determined by its reflected spectrum, which is heavily determined by the emission spectrum of the light source. Different white light sources can vary dramatically in their emitted spectrums with peaks in the yellow, green or blue bands rather than flat across the visible spectrum. The emission spectrum and intensity from a light source also changes as the light ages, or in response to temperature changes and other operating conditions. While monochrome systems face these same challenges, changes in lighting that would make a visible impact on a color vision system may not even be noticeable during a comparison of the monochrome versions of the same images.
Slight color changes can even occur after acquisition. According to product manager for color vision systems at Cognex (Natick, MA), Bashar Marshal, end users need to be aware of the consistency of their image processing software. ‘‘You need to be aware of the consistency of your image analysis software. When people build algorithms, you want to make them fast, but some times you can make them so fast that you sacrifice thoroughness. Getting a bad result fast doesn’t help anybody,’‘ Mashal said.
Applications in color space
As with monochrome machine vision systems, the application determines algorithm selection, but with a new twist: instead of those algorithms operating in a single continuum that stretches between white and black, end users have several mathematic representations of color or ‘color spaces’ to choose from. Color cameras have several standard color output signals, including Red, Green, Blue; composite color – which is similar to multiplexed Hue, Saturation, Intensity (HSI); broadcast television’s YUV standard; and raw data from single chip color cameras with built in mosaic filters. (According to Matrox Imaging’s (Dorval, Quebec, Canada) product line manager, Pierantonio Boriero, raw data streams are becoming more prevalent with the easy implementation of Bayer reconstruction filters.) Each of these output signal types is generally associated with a similar color space, but the available color schemes do not stop there. Another common color space used in the printing industry, and therefore, the machine vision industry is LAB color space, which identifies colors based on the response of the human eye to visible light.
However, while color cameras offer several different color outputs, three-channel chip color CCD cameras collect color information in only one way: RGB, explains DVT Sensors applications engineer, Eduardo Arcis. RGB values are then transformed into the appropriate color space coordinate. ‘‘One color space does not necessarily mean you getting better color content than another color space,’‘ Arcis said. ‘‘It’s mainly an issue related to different markets and industries.’‘
Color applications tend to fall into two broad categories regardless of industry, namely the detection of objects with clearly different colors, and the other is an objective color measurement that determines fine shade differences between two objects. These objects can be anything from automobile car mirrors and a matching door, to a golden template or proof and a stack of printed labels, for example.
Color, context and calibration
‘‘Low cost color vision sensors can tell you how far a color is off, or if an object is red. Then you have our category where you have a sensitive color system that can look at a homogenous or non-homogenous region and find fine color variations, patterns, things of that nature,’‘ said Bud Patel, director of marketing and business development for Applied Vision Company (Akron, OH). Applications requiring fine shade differentiation require more than just RGB. These systems usually combine RGB and HSI, or go some other six-plane color space, such as LAB or Delta E LAB. Some of the most advanced color inspection systems, such as those from Applied Vision Company and Advanced Vision Technology (AVT Inc., Hod Hasharon, Israel), convert RGB to LAB values and Delta E values. And through custom software, translate those color changes to specific suggestions for press operators.
AVT LTD.’s PrintVision Jupiter (left) automated print inspection system with remote optical heads (right) is designed to accommodate the physical dimensions of today’s large scale printing presses and to convert color changes determined in LAB color space into operational changes for the press operator such as plate pressure or ink mixtures.
‘‘Humidity, air temperature, ink batches and substrate batches all have a major role in color appearance,’‘ said AVT’s president and CEO, Shlomo Amir. ‘‘LAB is a good way to give feedback to the press operator to calibrate the machine. This technology will enable us to give information on ink balance or plate pressures or ink viscosity.’‘
DVT has created a different approach to fine color differentiation, combining a spectrograph with an automated imaging system. The DVT Spectral Camera takes reflected light from the object under test and passes it through a prism, which separates the light into a continuous spectrum. A 640 x 480 BW CCD sensor determines the color spectrum based on intensity distribution across the sensor. ‘‘DVT’s Spectral Camera is even better at detecting slight color variations than our color sensor, which can detect 16 million discreet colors. This gives you the ability to distinguish colors that are very close together in terms of green or blue content, for example,’‘ Arcis explained.
Because the environment housing the application (light, air quality, temperature, etc.) directly effects how a color appears, automated color vision systems have to constantly evaluate changing conditions to make color measurements as objective as possible. Calibration is one critical dimension to this process, but companies also use smart operating procedures to minimize the impact of changing contextual conditions.
For instance, Arcis explains that DVT color-CCD based systems use a two-step calibration process: dark image calibration and white reference. During dark image calibration, a ‘dark’ image is taken (usually with the lens cap fixed in place) and accumulated electrical current in the CCD chips is removed by filtering out pixels values that are not black. During a white balance operation, a ‘white’ reference object is imaged by the camera used to automatically adjust the color balance from the three chips.
By using self-learning algorithms and insuring that a white reference object is located in the camera’s view, DVT’s Spectral Cameras automatically adjust to changing lighting conditions during operation so that comparisons between an object under test and a stored digital proof can be normalized in the same color content and a successful comparison made even as ambient light or conditions change. Finally, documentation that clearly details all these procedures is the key to making sure that systems operating in different locations or at different times can produce the right result.
Applied Vision Corp. has developed a Digital Mastering Technology (DMT) that, in effect, includes a color reference in the stored digital template so that templates or proofs can be shared among various locations without concerns that transmission, viewing or other factors will impact the color quality of the proof.
AVT’s PrintVision Jupiter displays
Color comes of age
High resolution color inspection is another area that is starting to see growth. ‘‘Most of the requests we get are for defect detection on a color surface,’‘ commented Mark Sippel, product line marketing manager at Omron (Schaumburg, IL).
‘‘Not so much car doors, but small features such as the push button panels in a car door, or trim features -- things of that nature. And they need the ability to detect color and detect defects. Standard color cameras don’t cover the field of view they need with the minimum resolution to find the defects, so high resolution color cameras are getting interest…Color has been a pretty limited market – limited but growing. The number of applications that want color inspection is growing where it used to be consistent.’‘
Just as color vision systems have grown on the success of monochrome systems, color vision is also expected to benefit from anticipated improvements to color imaging technology. Faster speeds and flexible cameras with flexible color schemes are among the anticipated advances that will continue to expand the market for color vision systems, according to AVT’s Amir. ‘‘Area cameras that measure absolute LAB on top of RGG would really support a lot of these applications,’‘ Amir noted. ‘‘As line CCD cameras increase in speed from today’s levels of 8,000 lines per second upwards to four times that, we’ll see more applications in high speed press inspection materialize.’‘
There are currently no comments for this article.
Leave a Comment:
All fields are required, but only your name and comment will be visible (email addresses are kept confidential). Comments are moderated and will not appear immediately. Please no link dropping, no keywords or domains as names; do not spam, and please do not advertise.