Heterogeneous computing for vision

Identifies the computational requirements of tasks in vision. Vision tasks usually require three levels of processing/spl minus/low, intermediate and high/spl minus/each having distinctly different characteristics thus pointing to the need for heterogeneous computing. An example from the domain of photo-interpretation is used to illustrate some of the specifics of the processing levels. Finally, we discuss the pros and cons of using special-purpose and commercially available parallel computers for vision tasks.<
>