Color Plays a Role in Inspection

Color inspection systems now offer advantages over monochrome systems.

Mike Rogers, Innovate, Northampton, UK -- Test & Measurement World, 9/1/1998

Most inspection equipment uses monochrome (black-and-white) camera technology and gray-scale image-analysis software. But some inspection tasks can benefit from the use of a color inspection system. You might need to determine the color code on a component, for example, or match the color of a component to be sure it’s in the right location. Image-processing tools now provide sophisticated capabilities that let you search for or match colors to locate a specific component or feature. These tools can differentiate between subtle color differences a monochrome system can’t detect.

Until recently, the cost of color cameras and the cost of the computer time required to process color images put color inspection out of reach for most electronics applications. But the cost of color cameras has come down and more color-processing software has been introduced, so color inspection systems deserve consideration for difficult inspection tasks.

The case for a color inspection system becomes easier to sell to management because of four factors:

• Color cameras now provide capabilities and value comparable to those of monochrome cameras. Camera resolution should let you inspect 0201-size SMT components.
• Pentium-based PCs can quickly process the large amounts of data produced by color cameras. (Computer power is no longer a major issue in inspection systems.)
• Image-processing software now includes tools that can process and analyze color images.
• Lighting systems now provide stable sources of illumination necessary for proper color imaging.
  

Color inspection became important when people needed to recognize color bands on radial resistors. A gray-scale image of a radial resistor may show that the bands are different, but image-analysis software has great difficulty distinguishing the actual colors of the bands because

• some colors such as browns and reds are so close together on a gray scale that software cannot distinguish between them, and
• during manufacture, the application of the color bands isn’t always consistent, so resistors may have bands of inconsistent color quality.

To be fair, though, if you can obtain a gray-scale image that has sufficient contrast, the image-analysis software can perform good comparisons between a stored image and the object undergoing inspection. Even when the gray-scale values of two colors are so close as to make an inconsistent initial measurement, software can often enhance the contrast.

In Figure 1, you can see how a normalizing software tool can highlight the slightly different colors of the laser-etched markings on a device. Laser etching is not an exact science. Because the etched components have bits of burned-off materials on them, monochrome and color systems may have problems discerning and recognizing the etched characters. If you can inspect clean devices, however, a color inspection system can clearly differentiate the charcoal-gray plastic encapsulant from the yellow, brown, or orange markings produced by the laser.

Figure 1. The before (left) and after (right) photographs of markings etched onto an IC package show how software tools can extract useful information from indistinct images.

Don’t Merge Background and Components
A color inspection system also differentiates components from one another and from the background. SMT capacitors may arrive from a supplier one week as blue devices and the following week as brown. A monochrome system would find it difficult to differentiate between the blue and the brown capacitors on a PCB.

In addition, the PCB itself may appear at the same gray-scale value as either of these two colors or at least close enough so that the inspection system cannot differentiate the capacitors from the PCB. In this case, the system will produce false passes or false failures, depending on how the operator has it set up. Some monochrome inspection system suppliers have software “work arounds” that help systems differentiate between similar colors. The differentiation in a color image would make such a problem easier to solve.

In some applications, a monochrome system can confuse a large colored area, such as the top of a plastic connector, that has no clear edge or boundary in the image, with a similar area of a different color. For example, a monochrome system could have difficulty differentiating between a white connector and a pale yellow connector. A color system has an advantage when you must inspect these types of similarly colored products.

But the ability of a color system to detect differences between colors can lead to problems in itself. Although two colors may look exactly the same to your eyes and would pass your visual inspection for “sameness,” a color inspection system may evaluate them as different colors. Thus, a color inspection system must have some way to build in tolerance to slight color variations.

Here is how one company overcame the color-variation problem in its image-processing software. When Integral Vision (Farmington Hills, MI) used a conventional color-map scheme such as RGB (red-green-blue) or HSI (hue-saturation-intensity), the resulting map did not adequately define colors and allow for any tolerance of color variations. Experiments showed that for video-based color-vision processing, noise corrupted the color values produced in these color codes. The noise can come from several sources: the illumination sources, the camera’s electronics, the cables going from the camera to the frame grabber, and the frame grabber itself.

Although design techniques can minimize noise from these sources, they can’t eliminate noise completely. In a noisy environment, even a simple operation such as a test of color equality produces poor results. Assume that an inspection system acquires an image of a single-color patch and then calculates the average color of this patch. How many pixels exactly match the average color? The answer turns out to be very few pixels.

Instead of defining a color as a single value, Integral Vision defines a color as a volume in color space. Three parameters, one per axis, represent the average position of the color. A second set of three parameters provide standard deviations for the position on each axis. Thus, you end up with a small volume within the color space (Fig. 2). This type of representation assumes a Gaussian scatter for a given color, which experimentation shows to be accurate.

Figure 2. You can represent colors in a color space. But instead of letting a single point represent one color, standard deviations around a color point establish a color probability that aids in color matching.

When software uses six points to represent a color as described above, comparing two colors becomes a statistical operation. The software determines whether or not the statistical distribution that represents one color might have a good chance of representing the color you want to match.

Watch Lighting and Gain
Illumination also plays a vital role in color representation and thus in color measurement. Non-uniform illumination not only can distort the intensity distribution in a color image, but it also can significantly upset the color balance in an image. Similarly, different gain settings on the camera can wreck the color distribution completely. Your software and hardware must automatically and regularly compensate for these variations in lighting and camera gain.

You can probably get a uniform white illumination field over the target area and have all camera channels (one channel per color) exactly matched at setup. But the camera’s gains can drift, and the color temperature of the illumination source changes as the system ages. These problems also arise when you replace a camera or try to duplicate the existing system. Software suppliers offer several algorithms such as illumination compensation, dynamic white correction, and histogram equalization to help overcome gradual changes in the system’s characteristics.  T&MW

Mike Rogers is the managing director of Innovate, a manufacturer of test and inspection equipment that develops inspection systems. Mike worked for IBM, Plantronics, Zehntel, and Teradyne prior to helping launch Innovate in 1991.