Digital image processing tasks are highly computational intensive, because of the vast amount of data that must be processed, the real time character of many applications in this area as well as their complexity. This project proposes to carry out research on parallelisation of low and intermediate level vision applications. As a complement to externally funded projects, the following computationally demanding applications will be investigated in more detail: still and moving image compression, 3-D graphics, and simulation of parallel optical architectures for vision, which all suffer from severe speed limitations. A systematic approach based on a formal description of the image content and the analysis task will be followed to find optimal parallel solution strategies. Implementations on a connection machine, a transputer based parallel machine with a SPARC host workstation and simulated optical computing architectures will be compared. Optimally efficient parallel algorithms are not easy to design: the main difficulties which arise are (i) the degree in which the task is inherently sequential, (ii) the unbalanced load of the processors, (iii) the communication overhead. A research group on image processing and vision, such as ETRO-IRIS can only continue to produce relevant results, if it also considers and studies the algorithmic aspects of parallel image processing applications.

Runtime: 1996 - 1997