Parallel algorithms on compact binary objects
Thomas Johansson, Ewert W. Bengtsson · 2002
For SIMD computers, using virtual processors, a common strategy for mapping the processors on the image data is to apply one processor per pixel. For operations on gray level images it is a good approach, but for operations on binary images many of the available processors are idle and not used in the calculation. This paper presents some common binary operations on a new and more compact representation of the binary objects in an image, which leads to a more efficient processor utilisation and faster algorithms. The implementations are for a Connection Machine/200. The new data representation is tested on a typical application: the separation of touching objects.