Video-restoration software migrates from multiple DSP architectures

Several years ago, Texas Instruments (TI; Dallas, TX) announced that Dilip Krishnan and Showbhik Kalra, students at Singapore`s Nanyang Technological University (NTU), had won $100,000 in TI`s digital-signal-processor (DSP) Solutions Challenge, a contest that encourages engineering students to use DSPs for new applications. Under the direction of faculty adviser Chong Man Nang, associate professor at the university`s School of Applied Science, Krishnan and Kalra connected 15 TMS320C40 DSPs in a

Video-restoration software migrates from multiple DSP architectures

Several years ago, Texas Instruments (TI; Dallas, TX) announced that Dilip Krishnan and Showbhik Kalra, students at Singapore`s Nanyang Technological University (NTU), had won $100,000 in TI`s digital-signal-processor (DSP) Solutions Challenge, a contest that encourages engineering students to use DSPs for new applications. Under the direction of faculty adviser Chong Man Nang, associate professor at the university`s School of Applied Science, Krishnan and Kalra connected 15 TMS320C40 DSPs in a tree configuration during the development of a motion-picture-restoration system.

Because images generally do not change significantly from one frame to the next, frames preceding and succeeding a damaged image usually provide enough comparative information to detect degraded areas, to mathematically model the image region, and to fill in the degraded region with a restored image. "While most image-restoration algorithms blur images of moving objects in occluded or uncovered image regions," says Man Nang, "the Gaussian-weighted, bidirectional, autoregressive algorithm used in the system can recover lost signals in both covered and uncovered regions in an image sequence."

Although a close to linear speed-up was achieved using the DSP array, Man Nang and his colleagues have recently migrated the algorithms developed on the system to UNIX workstations from SGI (formerly Silicon Graphics Inc.; Mountain View, CA) due to user demand. The newly dubbed Revival motion-picture-restoration software runs on SGI O2 or Octane workstations running IRIX 6.2 or higher. It also allows restoration of images from various formats such those with a CCIR 601, PAL, NTSC, or HDTV resolution.

To digitize such images, SGI has recently introduced its DVLink, a PCI-based, IEEE-1394 add-in board for the O2 workstation. This board enables the workstation to transfer multiple digital video streams from other IEEE-1394-enabled video sources. Once digitized, the Revival menu-driven software guides the user to restore any specified set of segments of the motion picture.

"For example," says Man Nang, "a motion picture might consist of a few segments that require restoration work. So, we allow the user to specify the extent of the segment by entering the start and end frames or the in-and-out time code." After the images are processed, they can then be stored in a digital video-tape recorder or a Betacam videocassette recorder.

More in Boards & Software