SEAC and the Start of Image Processing at the National Bureau of Standards

Consequences of Early Engineering Decisions

Several decisions made in the construction of the first scanner have been enshrined in engineering practice ever since. Perhaps the most insidious was the decision to use square pixels. Simple engineering considerations and the limited memory capacity of SEAC dictated that the scanner represent images as rectangular arrays of size 176 x 176 square binary pixels, each of size 0.25 mm x 0.25 mm. No attempt was made to predicate the digitization protocol on the nature of the image. Every image was made to fit the Procrustean requirement of the scanner. That this was unnecessarily restrictive would have been obvious to the researchers had they known about the methods used 1400 years previously by the artists who made mosaics in Ravenna, Italy (12). The practice that NBS initiated with the SEAC scanner survives today, much to the disadvantage of economical storage of images.

Binary thinking was quickly supplanted as research progressed in the pattern recognition field. Over a period of many years, attempts to design vision systems have successively expanded image representation to include multiple gray levels, color, texture, motion, and stereo. Today, researchers in vision systems even realize that a unimodal approach to Image Understanding Systems is unnecessarily limited. Other sensory modalities like touch must supplant vision if image understanding is to be achieved.

One peculiar consequence of early decisions never occurred. In 1957 computer storage capacity was very expensive compared to computation speed. The SEAC had the equivalent of 6,000 bytes of storage. The basic operation time for a simple addition operation was 48 microseconds. It would appear logical to have investigated image compression algorithms for economical storage of images. But, in fact, the widespread development of such algorithms did not occur until recent years when megabytes of memory and gigabytes of non-volatile storage and operation times of a few nanoseconds were common. With storage so expensive in 1957, it was unrealistic to consider serious storage of large quantities of images. Only when storage became cheap compared to computation time was it reasonable to look for even cheaper methods of storage, hence image compression.

<< previous next >>