Dynamic programming for region based pattern recognition

We use dynamic programming to recognize patterns whatever the position of the objects in the scene, their orientations and scale. The program deals with a real picture, extracts regions, encodes and transmits them to a dynamic programming procedure (EM). The latter compares them to a form dictionary encoded through the same principle. This procedure has been applied to various dictionaries of patterns belonging to different applications. It proves to offer both, the flexibility of "elastic matching" even in the case of overlapping objects and the rotation and scale - invariance as a result of coding. The computational requirements are fairly low, all the more so as a parallel implementation would be easy.