Termination condition for a parallel shape coding and a parallel shape coding algorithm

Abstract

This paper presents a parallel shape analysis depending on simultaneous merging of elementary features detected through calculation o f the number o f objects within neighboring windows. Termination condition for parallel shape processing is formulated and proved. Hierachy of neighborhood operations for a parallel feature merging process is established to recognize more primitive features first (as tips and object bounds), then curved and straight strips, then regions, and finally forks and junctions. This hierarchy is necessary in order to have a complete data [feature) set adequate for consideration by the termination condition at a current recognition level. The parallel algorithm can be embedded on a SIMD grid architecture. Elementary image features are detected inside partly overlapping windows fixed in an image plane. Each window contents is processed by a separate processing element (PE). Two neighboring feature chunks are merged by adjacent PES, and the merged feature chunks are joined in each next Zth parallel step by every 2'th PE possessing complementary feature. The shape coding terminates i f a complete set of features (such as tips, bounds, corners, segments, ing t o the cell. The time complexity for such computations is O(number-of-curves). A search is needed t o match endpoints of strips found in adjacent subfields.The speed of our SlMD architecture is less sensitive t o the number of lines found in the image plane when compared t o a pyramid. The size o f a field (the number of direct dependents) analysed by a higher level cell is reduced by all adjacent object bounds detected outside of the field. The merging of these adjacent feature bounds (and o f adjacent more primitive features) will be performed by lower level cells.In our approach, not only feature positions are manipulated, but also the feature names. Elementary features are detected in a window using the numbers of window objects [16][17][18]. Different combinations of the numbers are characteristic t o specific image features (as segment, fork, tip, junction, discontinuity).For instance, the name "segment" is achieved if the number