IMPROVED OBJECT BOUNDARY REPRESENTATION BY INVESTIGATING NEIGHBOR QUADRANTS IN A QUADTREE STRUCTURE OF AN IMAGE.

A quadtree representation of an image enables efficient storage and approximate structural description of constituent patterns. A disadvantage of the method is the arbitrariness of dividing lines. Picture areas and the dividing lines imposed may combine to slice single objects into fragments which could be 'non informative. ' As a result objects present in the original image are unrecognizable or completely absent in the reduced picture. Search of neighbor areas of a quadrant should yield improved reduced pictures. A preprocessing predictive-correcting method is proposed. This decision procedure examines neighbor quadrants at any level of a quadtree and judges whether it is worth saving a quadrant or not by testing the surrounding area. The procedure is recursive based on three-level front-end algorithm for any pair of tree levels. A set of neighbor subquadrant templates are suggested for a better object boundary representation by determining neighbor subquadrant importance defined relative to parts of the examining quadrant. Other sets of templates could be used to detect Specific Shapes.