Whither the image computer?
Whither the image computer?
Andy Wilson Editor at Large
Over the years, the machine-vision and imaging-system industry has generated a range of products for the broadcast, computer, graphics, and multimedia industries. By endorsing the standards from these disparate industries, systems integrators have been able to configure machine-vision and imaging systems for industrial, scientific, medical, and military/aerospace applications. But for many companies that built early image-processing computers, the market was so fragmented that they abandoned their hardware offerings and changed direction into more lucrative areas such as motion-picture and video-game-related products.
For integrators of machine-vision systems, the diversity of standards can be confusing. Worse, rather than endorse any standards that might be available, some machine-vision companies are moving in the opposite direction, adopting proprietary camera interfaces and networking capabilities. Indeed, the lack of any major machine image-processing vendor has led OEMs into the domains of Microsoft Corp. and Intel Corp. These two companies have attempted to commercialize the vision industry by offering "standard" imaging-software-development kits and MMX processors.
All of these strategies have distracted the aim of imaging-systems design--to build a system that can easily inspect, diagnose, and make decisions on image data. However, there is no imaging system that can identify an image and automatically make intelligent decisions. For example, why can`t machine-vision systems automatically identify labels, determine whether they are damaged, and analyze their color and position without having to be "trained" first? What`s more, why can`t medical-imaging systems automatically identify objects in x-ray images and perform a diagnosis the way a physician does?
At present, systems integrators are moving away from C-callable libraries; instead, they are choosing to use graphical development tools. Certainly, this speeds the development process. But developers would really like to have machines that understand typical classes of machine-vision problems and adjust them for their specific application. So, what`s needed to accomplish this? Certainly, the hardware, networking, and computing capabilities are available. But no one company is currently offering an "image computer" that can handle the problems of scientific, medical, and industrial applications.
In the future, however, the standards now being driven by consumer, networking, software, and broadcast vendors will eventually force a model on image processing as we know it. Just as the multisync monitor turned the monitor market into a group of companies with products that were difficult to distinguish from one another, this imaging model will encompass image compression, multisync cameras, image-processing software, and networking.
Already signs of this model are emerging with Intel`s planned introduction of the Katmai processor, a 500-MHz processor slated for introduction in 1999 that will include MPEG-2 compression integrated on the 2-GFLOP processor. As this trend continues, mother board vendors are likely to enhance their offerings with digital-camera interfaces and real-time video-display capability. The result will be an "imaging" or "vision" computer that can be deployed in numerous applications at low cost. When this happens, don`t expect Intel Corp., Dell Computer Corp., IBM Corp., or other Fortune 500 companies to sit still. Finally, the fragmented image-processing market will consolidate, and the "imaging computer" will have come of age.