By Silvia Stoll, Messe Stuttgart
Monochrome cameras still dominate machine vision, but color is growing in popularity. "Five years ago the proportion of color area-scan cameras was around 10% and has continued to grow," explains Joachim Linkemann, VC product manager at Basler Vision Technologies (Ahrensburg, Germany;www.basler-vc.com). "In our company the ratio of color to black-and-white applications is currently around 1:3."
Linkemann explains that the biggest changes have been seen in linescan cameras, the proportion of which were negligible only five years ago. "Color-image capturing accounts for as much as 90% of our business," emphasizes Markus Schnitzlein, managing director of Chromasens (Konstanz Germany;www.chromasens.de).
Color filters
Today's color cameras feature common photosensors-- CCD or CMOS technology as used in black-and-white cameras. Depending on quality and application, the market offers 3-CCD color cameras featuring a sensor for each of the primary colors: red, green, and blue (RGB). In this design, the beam of light is split using a prism.
Single-chip cameras fitted with a color filter are considerably more common. A color filter in one of the three primary colors is positioned over every photocell. At this point, one pixel can provide information on only a single color component. Three pixels are then interpolated to form a picture element, and these are used to create the full image. The true resolution is therefore lower than the number of sensor pixels. The type of filter used for the sensors varies according to manufacturer, but the Bayer filter is generally preferred. This filter uses a group of four pixels for each picture element: two green, one red, and one blue. The majority of camera manufacturers use the same sensor, yet "market differentiation is defined by the quality of electronics, which are characterized by low-level 'image noise'," explains Linkemann.
"I see the major challenge in demosaicing," adds Gerhard Holst, head of research-and-development at PCO AG (Kelheim, Germany;www.pco.de). "How do I generate a color-precise final image using information on single colors? This can be a highly costly and time-intensive process. Last year we invested a great deal of time into this and we can keep pace with--maybe even surpass--the best color algorithms published in the scientific world." PCO AG is currently working on a new CMOS high-speed camera.
Illumination is the challenge
Linkemann identifies the main challenge in the chain of different machine-vision disciplines, however: "If the illumination is not right, you won't achieve satisfactory results, no matter how good your algorithm is. This means that illumination, imaging optics, and the camera itself first have to capture a reasonable picture before data transport and image analysis can take place."
He explains that objectives are frequently not good enough or not selected with enough care to achieve the required physical resolution. According to Linkemann, "Every euro saved in light, optics, or the camera itself is subsequently lost again several times over in the software."
The level of illumination plays a crucial role in color-image capture. On the one hand, the "sensitivity of a sensor is substantially reduced by color filters by around 20% to 30%," states Holst. "This means that you often need more light than you have at your disposal."
Continues Markus Schnitzlein of Chromasens: "On the other hand, a light source should deliver a uniform prismatic white light as far as possible if you're dealing with high-quality color imaging or even colorimetry, which depends on obtaining a color-precise picture." This could be the case when inspecting wafers or archiving paintings. "We can use specially designed white LEDs or xenon-based high-pressure lamps for this purpose," explains Schnitzlein.
Chromasens favors self-developed, high-quality color linescan cameras. "Compared with an area-scan camera, they have the advantage of coping with much higher data rates and very large scan-width resolutions," continues Schnitzlein.
A requirement when using linescan cameras, however, is that 'something moves' (either the object, the web material, or the camera), otherwise the same line would be recorded repeatedly. Each of the three primary colors occupies a sensor line in the case of a trilinear linescan camera (single-chip camera). "Our sensor units are characterized by an impressive line length, which can reach up to 22,500 pixels per line," adds Schnitzlein.
"The speed is also important. We make cameras with 7500 pixels and 3 (RGB) x 60 Mbits/s. That's a fairly unique achievement." But when are such performance levels required in practice? "If you want to scan objects or pictures with a width of 1 m and a resolution of 40 μm (approx. 600 dpi), you will quickly make full use of the 22,500 pixels," continues Schnitzlein. He adds that in the case of metrological activities, three color channels would not be sufficient, however. Here you would use between six and eight, or sometimes up to 16, color channels.
Chromasens also has tackled the problem of lamp deterioration. "When we scan an image, we also measure the spectrum of the light source and can subsequently correct the color values directly," says Schnitzlein. "All our color cameras are equipped with a continuous white-balance function. And since there are so few sophisticated lighting systems on the market, we will be offering our own light sources with our camera systems. We see a big potential for increasing stability and quality in this area."
Color precision
"In many areas of industrial imaging, color quality is becoming increasingly stringent," states Andreas Schaarschmidt, manager of sales and marketing at Stemmer Imaging (Puchheim, Germany; www.stemmer-imaging-group.com). "This means that the demand for 3-chip cameras will continue to rise." Expert opinions differ on this subject, however: "Three-chip cameras are considerably more expensive and require special objectives," argues Linkemann. "With linescan cameras, there is a clear trend towards 1-chip cameras; 3-chip cameras are preferable only in the case of falling, rotating, or nonlinear moving objects."
"Three-chip cameras are indispensable in applications demanding extremely precise color reproduction, such as banknote inspection," suggests Ingo Lewerendt, product manager at Allied Vision Technologies (Stadtroda, Germany;www.alliedvisiontec.com). "However, the disadvantages are the elevated price and the larger dimensions, which complicate integration in compact systems. The market share of 1-chip cameras continues to rise. This is also due to the fact that most industrial applications do not necessarily require the color precision offered by the 3-chip solution. What is more, color sensors featuring increasing resolutions are entering the market, which means that the color precision offered by 1-CCD cameras is closing the gap on 3-CCD cameras."
Compared with 3-CCD linescan cameras, trilinear linescan cameras have the disadvantage that they view different aspects of an object for each color channel. This causes a displacement (shift) measured in microns, which can be corrected on occasion, but particularly when changes to local illumination occur (through buckling or waves on a moved web material), color deviations will become apparent. These deviations can often only be ironed out using a highly elaborate synchronization process.
Using 3-CCD cameras, on the other hand, means that brightness changes can be seen in all three channels simultaneously and these are not displaced in terms of time or space. "With 3-chip cameras, the colored pixels (RGB) are situated precisely on the same optical axis," explains Schaarschmidt. "This generates a considerably improved color quality for the viewed object." JAI offers a highly compact 3-CCD prism camera, which includes three sensors with 2048 pixels in combination with a special prism. "This camera delivers outstanding color resolution and fidelity and achieves a maximum 19-kHz line rate," says Schaarschmidt.
Despite these apparent drawbacks, trilinear color linescan cameras are undergoing development. The RGB linewidth in this new generation is said to be 30 μm, which ensures that a line-shift problem would be balanced with a minimum of effort using an internal software-correction procedure.
More applications
Color cameras can now be used for a broad spectrum of application areas. "These include the printing industry," says Schaarschmidt, "in which surfaces and print quality is subject to close inspection. And color systems are used in the food industry for recognizing and differentiating fruit and vegetables, for example. In medical technology, color image systems are applied in microscopy and dermatology for the purpose of identifying skin diseases. But also in the classical industrial environments of mechanical engineering and plant construction, color imaging can make an important contribution, for example, in the auto industry (inspecting paint surfaces and leather upholstery) and the timber industry (classification of wood materials)."
"It is often the case, however," qualifies Linkemann, "that dedicated image processing is still monochrome yet viewers and operators are increasingly demanding colored images for visualization." He identifies a trend toward rising sensor resolutions. However, since machine-vision data also need to be processed, the processing speeds for these resolutions are not even close to matching those seen in the consumer-goods sector, though these are predicted to rise. At the same time, cameras are set to become smaller, more powerful, and more affordable. And future cameras will feature preprocessing and image-correction functions.
The trend toward multispectral cameras cannot be overlooked. TVI Vision (Helsinki, Finland;www.tvivision.com) recently added to its product portfolio a 3-CCD linescan camera that is also sensitive to the near-infrared (near-IR) spectrum. "This multispectral camera includes a specially configured synchronized beamsplitter that separates the spectrum into blue, green, red, and near-IR and directs these to the three line sensors," explains Schaarschmidt. He adds that the trend toward significantly higher resolutions on the basis of CMOS sensors will also soon enter the market: "But above all, new products in the color linescan camera segment will generate new solutions and promote new ideas. Of course, the GigE Vision Control Protocol will play an increasingly significant role in achieving 'simple integration.' And, thanks to the popularity of HDTV television (1080 standard), available sensors, DSPs, and prism technologies will enter the market and also be used in the industrial camera market at more and more competitive prices."
At the VISION 2007 trade fair (Stuttgart, Germany; November 6—8, 2007), the trend toward greater use of color will become apparent. VISION is the international platform for industrial image-processing and recognition systems in Europe, and this year it will present for the 20th time a combination of the latest innovations, products, and specialist knowledge from the world's leading machine vision providers.
Silvia Stoll can be reached at tel. +49 711 2589-696, Fax ext. 305, or e-mail:[email protected].
