Image compression: Ready for prime time?

As the application of machine vision grows for inspecting electronic products during manufacturing, demand increases for better ways to archive and transmit the resulting images more efficiently. Companies often want a permanent record of virtually every image generated by their processes for troubleshooting and as protection against lawsuits. The ability to store more images and transmit them more easily from one location to another has become imperative.

Although compression techniques—most notably from the Joint Photographic Experts Group (JPEG; www.jpeg.org)—have been available for many years, electronics manufacturers have not embraced them for fear that compression will degrade image quality. A new version of the JPEG standard—dubbed JPEG 2000—offers to improve compression ratios without sacrificing quality. Will this new approach finally entice the manufacturing world onto the bandwagon?

For electronics manufacturers, the two most important issues in vision-based inspection are image resolution and real-time acquisition speed. The conventional JPEG compression technique can shrink a file without loss of information, achieving a relatively modest compression ratio of 2:1. JPEG 2000 increases that ratio to as much as 4:1, substantially reducing file storage and transmission requirements. Electronics manufacturers will likely not adopt higher compression ratios that do lose information.

Lossless JPEG compression begins with a predictive-coding step. Predictive coding depends on the fact that in any image, most adjacent pixels share similar values, so an algorithm can predict a pixel's value from the value of its neighbors. Storing only the differences between pixels requires fewer bits than does storing their full value. A subsequent encoding step assigns a variable length "code word" to each pixel based on the number of bits by which adjacent values differ.

Unfortunately, JPEG 2000 achieves higher compression ratios by using more computationally intensive algorithms. As a result, a software-only JPEG 2000 implementation cannot keep up with real-time manufacturing requirements. According to Pierantonio Boriero, product-line manager at Matrox Imaging (Dorval, QC, Canada; www.matrox.com/imaging), real-time JPEG 2000 compression would require computers that run at more than 30 GHz—an order of magnitude faster than today's 3-GHz machines.

Fortunately, solutions are emerging. Matrox's own Morphis frame grabber overcomes the computational limitations by including a real-time hardware compression and decompression accelerator. A large, dedicated buffer provides reliable image capture to the onboard accelerator or the host PC.

The question remains whether electronics manufacturers can overcome their historical reluctance and embrace compression at all. The perception that compression will inevitably lose information represents a formidable barrier. Also, ever-increasing storage capacities and transmission bandwidths reduce the disadvantages of uncompressed images. Nevertheless, if manufacturers truly want to archive all pertinent data, image compression will have to represent part of the equation.