Technology Improvements Boost Color Applications

A few years ago, color vision systems with price tags in $50,000+ range were not that uncommon, but customers willing to pay that much were. Today, smaller dedicated systems that use PC hosts or smaller numbers of dedicated processors rather than groups of processing boards are opening up color applications in the automotive industry as well as web inspection, printing, plastics, electronics, transportation, pharmaceutical, food and packaging among others.

More than just cost considerations, these new color systems are finding ways to handle the larger color data sets in real time using software that resembles or even straddles the line between color and grayscale image processing. These color vision systems can detect slight color variations as well as locate surface defects – all the while limiting the detrimental effects of lighting, depth and shadows on color image processing systems.

Categorizing color vision systems
Color vision systems fall into three basic categories, although systems that are more complex may fill two or even all three categories. Those categories include: systems that detect the presence or absence of a particular color, but are not particularly well suited to differentiating between two very close colors; systems that can detect presence/absence of colors and perform many traditional gray scale image processing spatial functions (edge detection, thresholding, etc.); and specialized systems that use a variety of industry standard color schemes such as CIE, Lab, Delta E, and LCH (Hue Saturation Intensity (HIS)), or wavelength sensitive optics to make absolute color measurements.

Omron (Schaumburg, IL) is one company that is targeting entry-level color applications through its F400 series color image processing system. According to Mark Sippel, Omron’s vision product marketing manager for the Americas, color vision applications make up about 5 to 8 percent of their vision product base. Few of these applications require gauging or spatial measurements. Omron uses a combination of Hitachi (Tokyo) RISC processors and field programmable gate arrays (FPGA) to give its small footprint color system the horsepower to work in HIS space in real time. By identifying colors based on their hue, saturation and intensity, Sippel said that changes in illumination do not significantly change the hue and saturation values.

‘‘The largest group of applications for us is in automotive. We do a lot of paint differentiation. We also have sold the F400 into the packaging industry. Typically, we do feature inspection or gauging applications in gray scale, but when color detection is a must we go with the F400.’‘ When color and feature detection are important, Omron tend to use a color and gray scale system, or pushes the color application into gray scale, when possible, by using colored light and filters to isolate target colors within a gray scale image.

Keyence (Woodcliff Lake, NJ) takes another approach, straddling color and gray scale image processing to delivery a system with both color and defect identification abilities. In 1998, Keyence introduced a color binary CV301 vision system. The operator essentially identified a color, and the CV301 confirmed its absence or presence within the image frame. Last year, the Keyence introduced the CV700. The CV700 can accommodate multiple cameras at a relatively low cost by keeping the single color identification and reducing the remaining colors into gray scale, or a meshed color and gray scale approach, according to Keyence’s color technician, Josh Jelonek. By using this approach, the CV700 can use two cameras to look at the front and back of an object, and do defect identification using Keyence’s edge detection gray scale image processing algorithms in a relatively small package. ‘‘It’s less expensive to buy one system that can do color and location rather than one system for each,’‘ Jelonek said. In addition to automotive, the CV700 has found a home in the semiconductor, food packaging and plastic bottle manufacture.

Making the complex simple
Cognex (Natick, MA) uses 640x480 single chip color cameras in its Insight 1000C color image processing system. The 1000C works in RGB space in addition to HSI, mainly because of the camera. The camera includes front end processing with a mosaic RGB pattern on the chip to interpolate red, green and blue color values without sacrificing significant resolution. The RGB values are then converted to HSI to help eliminate the systems sensitivity to changes in ambient lighting, although Cognex’s product manager for color, Carl Gerst, said that RGB color values can be hardened against changes in illumination by focusing on two of the three color values that change the least in response to changes in illumination. The advent of white LEDs has also improved the performance of color systems for automated inspection. Not only is their spectral response better across the visible spectrum, but also these systems are more stable flash to flash than xenon or other strobes and use less power.

According to Gerst, the 1000C offers many of the same functions as a grey scale system, but is limited in speed and to a lesser extent resolution. ‘‘The 1000C does not operate at frame rate. The 1000C can inspect 3 to 4 parts per second, which is still pretty fast,’‘ he said. Cognex’s older system, the 900C with a three-chip CCD camera and more processing power fills the slot for applications with higher demands.

Bringing greater functionality to less expensive systems, however, is the goal. Cognex’s 1000C offers an integrated run-time and development environment that makes setting up the system easier. An integrated spreadsheet program also makes data evaluation, setup and evaluating system performance easier. ‘‘That’s what you’re going to see us addressing in the future, reducing the cost of buying the system, getting it set up and qualified, and operating it,’‘ Gerst said.

Like Cognex, DVT (Norcross, GA) also focuses on making color vision simpler to operate. ‘‘We had a large application base that knew how to work with gray scale systems and how to solve applications with those processing tools, but they needed the capability to recognize color…So we added the capability to learn the color and use our gray scale tools that customers recognized,’‘ explained Ali Zadeh, senior R&D engineer at DVT. 

Using a familiar front end, DVT has added three major features: clustering, matte areas and wavelength sensitive optics. According to a Zadeh, clustering is a type of thresholding that identifies all the acceptable variations of a color for a given object. By learning all acceptable colors of an object with the 542C system at different angles and under different illuminations, the system determines the shades of acceptable colors in RGB space. This ‘cluster’ of 3D color thresholds is then combined into one data set and the complete color spectrum is reduced to binary values that DVT’s gray scale algorithms can accept. This greatly reduces the amount of processing, but still allows for fluctuations in lighting and object presentation. ‘‘The learning curve is much shorter because for the customer, it’s just another thresholding method,’‘ Zadeh said.

‘‘Clustering an Orange Object’‘

Matte areas also help to reduce the potentially detrimental effects of illumination. With the matte area function, an operator identifies the appropriate color for a part and then a line is drawn through 3D space from the color back to black. With this feature, the system accepts colors that fall within a certain distance in RGB space from the matte line, making the assumption that a reduction or increase in illumination is the cause for color’s variation from the reference color.

For the most demanding color applications where small color variations are critical, such as individual parts in an automobile door where one model may have several variations of white, DVT has developed the SpectroCam. The SpectroCam uses a Spectrograph to split the incoming light into its constituent wavelengths along the X axis and spatially along the Y axis. With this product, the complete spectrum (380nm – 900nm) of the sample can be learned and small deviations detected. The peak location of the spectrum can also be monitored for LED quality control. The system is highly sensitive to color, but works best when the object under test is uniform in color.

Advanced color controls
When color and feature identification are critical at high speeds, Applied Vision Compnay LLC (Akron, OH) leans on the full power of today’s PCs. Applied Vision combines template matching – a more complex version of pattern matching that compares pixel to pixel to generate a confidence match value across the entire area – with frame by frame normalization for illumination differences. ‘‘Uniformity and light intensity corrections are critical to color. We can compensate for variations in lighting and look for small color shifts and defects and capture them,’‘ said Dr. Richard Sones, chief technical officer at Applied Vision. ‘‘We can look at a tile with many colors and give you a Lab [value], Delta E difference, RGB or average color variance for that area…You have to be very fussy about calibration to do this. To do these massive calculations you need the newest PCs. Multiple gigahertz processors now make this sort of thing possible at physical line speeds…Massive memory is also required for a large number of image buffers. With a 1 MB color image at standard VGA you need hundreds of megabytes for color corrections with high bus speeds.’‘

Applied Vision has sold their system to inspect both sides of credit cards at 500 pieces per minute, ‘‘…and it’s only in the last couple of years that you had a PC capable of doing that,’‘ Sones said.

This kind of processing power will continue to filter down from the most powerful to the entry-level systems, experts predict. Today’s high-clock speed microprocessors are enabling color systems with megapixel resolutions, 3-chip cameras and multiple cameras, reducing the overall cost per camera by sharing the processing costs among many sensors. Digital color cameras are also expected to improve, moving from progressive interlace to high-speed readout cameras, preferably with a consolidated move towards either 1394 or Camera Link digital standards.

Other anticipated enhancements could come from a move away from the Asian-preference for RS232 to the more rugged Ethernet standard and CMOS sensors with additional front-end color processing and normalization capabilities. ‘‘I see distributed processing systems really making a greater difference than the incremental increases in microprocessor power,’‘ said Omron’s Sippel.

Companies Mentioned:

Applied Vision Company, LLC
95 Hanna Parkway
Akron, Ohio 44319
Telephone: 330-724 9600
Fax: 330-724 4154

Keyence Corporation of America
50 Tice Blvd
Woodcliff Lake, NJ 07677
201-930-0100

Omron Management Center of America, Inc. (OMCA)
Regional Management Centre
1300 Basswood, Suite 100 Schaumburg, Illinois 60173 U.S.A.

Cognex Corporation
One Vision Drive
Natick, MA 01760-2059
Phone: (508) 650-3000

DVT Corporation
1670 Oakbrook Drive, Suite 330
Norcross, Ga. 30093
Phone: (770) 449-4960   Fax: (770) 449-3073

Hitachi LTD
6, Kanda-Surugadai 4-chome, Chiyoda-ku
Tokyo 101-8010, Japan
Phone: +81-3-3258-1111
Fax: +81-3-3258-2375