A method for the hierarchical extraction and description of lines and solid regions in technical drawings

Many papers dealing with technical document analysis claim that the greatest problems come from the numerous and very different levels of representation required to finally interpret the pixel image of a document. These difficulties raise general questions like: what kind of information has to be extracted and at what stage of the process? How should we encode the information acquired after each stage? To solve these problems, we try to integrate natural techniques normally associated with human perception and have thus perfected a method of describing technical drawings using only one model of representation. It must allow us to depict all the shapes from the original document at the different levels of analysis as clearly as possible. We have chosen to accomplish this description with a limited number of simple primitives: from the pixels base, we first build vectors and then quadrilaterals. This shape polygonization method seems to be adapted to the different kinds of objects found in such technical drawing images and, actually computation time is minimised.