Binary image representation and coding by a double-recursive morphological algorithm

摘要

This paper presents a recursive morphological operation which can be advantageously substituted to classical erosion with an application for image coding. A structuring element is translated over binary image following four scanning modes. The result of the operation, as for classical erosion, corresponds to the centers of the structuring elements which are included in the image objects, but our operation authorizes only a minimal overlap between the translated structuring elements. This minimal overlap needs to take into account previous erosions, so the operation is recursive. It results in a subset of the erosion, containing fewer points than the erosion, but gives a good approximation to image objects. In a second step, operation is extended with a second recursivity level corresponding to structuring element size variation. This new double-recursive morphological algorithm describes an image object, from coarser structures to finer details, by the loci and sizes of the translated structuring elements which are included in the object. This algorithm provides a minimal redundancy shape representation. To evaluate this algorithm in the context of binary image coding, we propose two different techniques for encoding representative points issued from our shape decomposition algorithm. Comparative experimental results with two classical methods (Modified Huffman RLC and TUH code) and also a morphological method (discrete skeleton), expressed in terms of compression ratio, show the good performances of our approach.