The PCI bus speeds up image-data handling
During the past year, what might be called the "battle of the buses" was under way in the image-processing arena. That conflict, which involved the ISA bus, frame grabbers with local memory and processors, and 486-based systems on one side and the peripheral component interconnect (PCI) bus, memoryless frame grabbers, Pentium processor, and Windows NT on the other, is over. In fact, it was never actually a battle, but rather the phaseout of an aging technology by its successor. The PCI, Pentium
The PCI bus speeds up image-data handling
Adoption of the peripheral component interconnect (PCI) bus permits digital cameras,
full-color imaging, and versatile frame grabbers to capture new applications.
James V. Flatten
During the past year, what might be called the "battle of the buses" was under way in the image-processing arena. That conflict, which involved the ISA bus, frame grabbers with local memory and processors, and 486-based systems on one side and the peripheral component interconnect (PCI) bus, memoryless frame grabbers, Pentium processor, and Windows NT on the other, is over. In fact, it was never actually a battle, but rather the phaseout of an aging technology by its successor. The PCI, Pentium, and Windows NT combination made PC-host-based imaging an attractive alternative to frame grabbers with local memory and processors, even for the purchase of completely new systems.
The technological argument was simple--the ISA bus, with a 16-bit data path and effective throughput rates of about 2 Mbyte/s, cannot transfer video in real time, which requires a minimum throughput of about 10 Mbyte/s. The 32-bit PCI bus, with a capability of 90 Mbyte/s, coupled with a Pentium processor and the 32-bit Windows NT operating system, provided a solution for real-time video. High-end image-processing applications were now possible on a PC; dedicated processors were no longer required. The PCI bus also eliminated the need for the frame grabber to have both onboard memory and graphics, as well as circuitry to connect to a digital-signal-processor (DSP) or other onboard processor.
Going with the flow
The PCI bus renders onboard memory unnecessary because image data are transferred across the bus as they are collected. On an ISA bus, image data are gathered faster than they can be transferred across the bus and must be temporarily stored in an onboard buffer memory before transfer to system memory. The PCI bus eliminates this memory redundancy and permits direct data transfer to the system memory or a display for imaging (see figure on p. 142) The large chunks of data from video cameras, which send the equivalent of 30 still images per second, result in data streams of 10-40 Mbyte/s. The 90-Mbyte/s transfer rate of the PCI bus is sufficient for live-video streams, eliminating the cost of onboard memory. Performance upgrades can be facilitated through system improvements such as increased memory or CPU speed.
For example, a typical DSP card equipped with an off-bus interconnect, a high-end processor, and 8 Mbyte of RAM costs about $8000. Add to that about $1500 for a 486 PC with an ISA bus and $2000 for a frame grabber with onboard memory for a total system cost of $11,500. In comparison, a Pentium PC with a PCI bus runs about $2500, and a frame grabber without onboard memory and graphics costs about $1000. Total cost is $3500, a net saving of about $8000.
ISA still vital
However, this PCI-Pentium-Windows NT combination is not the ultimate answer for all image processing. There is a large base of installed PCs with ISA buses using frame grabbers with onboard memory and processors. For many applications needing only low-frame-rate imaging, this technology serves more than adequately. Continued sales of high-performance frame grabbers attest to the popularity of ISA-bus machines. However, the market is assuredly moving away from ISA-bus to PCI-bus architectures. While most new PCI systems still come with ISA slots, these eventually will cease to exist in new machines.
But outboard-processor-based solutions will continue to play a valuable role in specific applications. The PCI-Pentium-NT solution is inherently not a real-time environment and, in demanding real-time applications, can only guarantee data capture for events that occur within a certain time span. For some critical applications involving very large am ounts of data or complex image-processing algorithms, true real-time results can be obtained only by assigning the processing tasks to an outboard-based processing engine.
For the majority of image-processing applications, however, frame grabbers with local memory and processors and the ISA bus are quickly approaching obsolescence. Prices for PCI-Pentium-NT systems will continue to fall, and upgrades for faster processing will further enhance performance and leverage investment in up-to-date technology.
The move to digital cameras
While still expensive compared with analog cameras, digital cameras offer strengths of flexibility, accuracy, speed, and ease of use. With camera vendors struggling to dominate the emerging digital market, competition will combine with economies of scale and advances in technology to drive down prices in the next two years.
When evaluating digital versus analog cameras for image-processing applications, it is important to remember that the processing must go through the same steps, whether in a camera, a DSP, or a host processor. In general, however, digital cameras have inherent advantages over their analog counterparts, mostly because there is no conversion circuitry. The elimination of this circuitry significantly improves performance and simplifies design, particularly in the following areas:
256+ gray levels of resolution. Technically, increased dynamic range has nothing to do with whether a camera has analog or digital output. However, digital-output cameras have become the de facto standard for 10- to 16-bit applications. Higher dynamic range is often required in x-ray and other medical applications, in astronomy, and in thermal imaging.
Lower noise. By digitizing the image at the charge-coupled-device (CCD) camera rather than at a frame grabber, signal noise is typically lower, resulting in better gray-scale image accuracy (higher signal-to-noise ratio). By its nature, analog video is susceptible to noise, whereas digital is not. In addition, by digitizing at the camera, potential signal degradation caused by internal PC noise or a poorly shielded analog input cable is eliminated.
Transmission length. Because of bandwidth limitations of analog video, the distance between a camera and the PC that processes the image capture is limited to roughly 2 to 3 m. Digital data, which use differential drivers, can be transmitted more than 10 m over an RS-422 cable without special requirements, making it ideal for use in harsh industrial environments.
Digital frame grabbers
Using a digital camera with a digital frame grabber creates additional benefits in high-performance applications. These advantages include speed for inspection and machine vision (8-bit processing is typical, with 128 ¥ 128 to 1024 ¥ 1024 spatial resolution), and accuracy for the image analysis market (12- to 16-bit processing, with 1024 ¥ 1024 and higher spatial resolution). Digital frame grabbing produces lower jitter, because there is none on the frame grabber; results are based on camera performance and higher speeds, up to 40 MHz. Other digital applications include thermal imaging (12-bit and higher processing), structural analysis, astronomy, ballistics and pyrotechnics, x-ray, traffic, and surveillance.
While commonly referred to as a frame grabber and performing the functions of one, a digital frame grabber is actually a board that functions as a digital-video interface. This type of product may appear to eliminate the need for a frame grabber, but a camera interface is still required to get the data into the PC and to provide camera control signals, such as exposure/integration, reset, horizontal and vertical synchronization, and pixel-clock input/output. The interface also provides for flexible image manipulation, including scaling, cropping, or region-of-interest selection, and for trigger and strobing control.
In terms of flexibility, users want maximums for bit depth, pixel-clock rate, spatial resolution, camera control capabilities, and dual-channel capacity. Accuracy is not relevant because analog/digital conversion is done in the camera. Digital frame grabbers can deliver these needs with high reliability and good price-for-performance.
Finally, output formats depend on the camera vendor, making it important to ensure that a frame grabber is able to work with the camera being used. Some cameras provide both analog and digital outputs--RS-170 is used for viewing and digital is used for processing. A digital-only frame grabber eliminates a modularity compromise. Frame grabbers that use a combination design may cost more and be less reliable because of additional parts and circuitry.
Another facet of the digital camera market is the evolution of CCD cameras and their emergence in image processing. The increased numbers of pixels in the cameras yield higher resolution, and, as more of the processing is shifted to the camera, output is digital instead of analog. Most digital cameras are based on CCD technology, but recent developments are pointing to the use of complementary metal oxide semiconductor (CMOS) sensors as a technology that, while more expensive today, will most likely become a viable, less-expensive alternative in two to three years (see Laser Focus World, Feb. 1997, p. 129).
The shift to color
The use of color frame grabbers is growing in machine vision and scientific imaging applications (see photo on p. 141).
Compared with monochrome, color requires three times the bandwidth--if 8-bit black-and-white image processing pushes an ISA bus, then 24- or 32-bit color will swamp it. The PCI bus has ample throughput to support true color at the NTSC- and PAL-format resolutions of 640 ¥ 480 and 768 ¥ 576 pixels, respectively.
Scientific imaging-analysis applications typically involve analysis of microscope-generated images such as stained cells, plant structure, or Petri-dish examination. Other uses include scientific visualization applications such as cartography, terrain reconstruction, medical-image enhancement, environmental-comparison studies, infrared thermography, medical diagnostics, pollution monitoring, weather forecasting, and personnel security identification.
Several image-processing trends are apparent. Windows NT will continue to grow in popularity, generating even more image-processing applications. And increasing standardization will result in the ability to easily upgrade both hardware and software.
Other trends to watch for in the next year or two include the emergence of digital cameras and frame grabbers. The move to color will continue. Manual operations will be replaced with automatic ones in applications such as image analysis of clinical diagnostic tests and fingerprint analysis.
Near-term applications of Intel`s recently released multimedia extension (MMX) technology for the Pentium chip will include acceleration of the inner loops of JPEG and MPEG compression algorithms. Use of MMX technology in scientific and machine-imaging applications is expected to follow. o
ISA-based frame grabbers store image data in an onboard memory buffer because images are gathered faster than they can be transferred across the ISA bus (top). PCI-based systems can send image data directly across a PCI bus for rapid transfer to the CPU or system memory (bottom).
JAMES V. FLATTEN is a senior product manager in the Imaging Group, Data Translation, 100 Locke Dr., Marlborough, MA 01752-1192.