This paper presents a method for vectorizing the graphical parts of paper-based line drawings. The method consists of separating the input binary image into layers of homogeneous thickness, skeletonizing each layer, segmenting the skeleton by a method based on random sampling, and simplifying the result. The segmentation method is robust with a best bound of 50 percent noise reached for indefinitely long primitives. Accurate estimation of the recognized vector's parameters is enabled by explicitly computing their feasibility domains. Theoretical performance analysis and expression of the complexity of the segmentation method are derived. Experimental results and comparisons with other vectorization systems are also provided.

In this paper, we discuss how the focus in document analysis in graphics recognition has moved from re-engineering problems to indexing and information retrieval. After a few reviews of ongoing work on these topics, we propose some challenges for the years to come.

We propose a method for indexing technical drawings. Our features are based on the notion of F-signature, which is a particular histogram of forces. The force histogram has low time complexity and describes a signature which is invariant to scaling, translation, symmetry and rotation. This article presents a new application of such signatures in the field of document analysis, and tests different characteristics of the F-signatures, like their sensitivity to the shape of the objects and to noise. Finally, experimental results show the effectiveness of such an approach. A brief overview of pattern recognition approaches dedicated to technical document indexing is given. The notions of histogram of forces and of its properties are presented.

In the context of graphics recognition, arc detection consist in the extraction of circles and arcs from the image of a graphics document or from the segments yielded by its vectorization. Several methods have been proposed for this purpose, and we briefly survey them. Then, we describe an improved algorithm inspired by two existing methods, and including a fitting step for better precision.

Vectorization, i.e. raster-to-vector conversion, is a central part of graphics recognition problems. We discuss the pros and the cons of basing one's vectorization process on skeletonization. While distance skeletons have proven to be robust and precise, they tend to distort the results at line extremities and junctions. In these cases, contour-matching approaches yield better results, but they have their own specific problems. A perspective is probably to combine the best of both methods.

We claim that time has come in graphics recognition for choosing stable and robust methods, even (or especially) when this means implementing methods proposed by others, instead of inventing a new algorithm which ends up being a minor variation on an old idea. In this spirit, we present some of the choices we have made. We do not cover the whole range of methods and steps used in graphics recognition, but rather concentrate on those which we deem the most crucial and time consuming, in terms of development time, in the design of a graphics recognition system. They include: binarization, text-graphics segmentation, thin-thick separation, vectorisation and arc recognition, and tiling.

Lately, our team has begun investigating the analysis of architectural drawings. The paper presents our first results. After a brief introduction to the specificities of architectural drawings and a short review of existing work, we describe the low level processing steps we have implemented: segmentation, vectorization and detection of arcs, loops detection. In the present state of our work, we have investigated two complementary techniques for higher level analysis; one is based on geometric analysis and symbol recognition, the other relies on the idea that architecture is a combination of spaces, and is therefore based on spatial analysis. We present our current results with these two techniques, and we suggest a number of perspectives for the continuation of this work.

Among the different aspects of converting an engineering drawing to CAD, 3-D reconstruction from several views has only been dealt with from a geometric perspective, although a lot of information is of symbolic nature and can be recognized as such directly on a single view. In this paper, we present a first prototype of a system combining geometric reconstruction through the "fleshing out projections" paradigm with symbolic recognition and matching.

Two kinds of knowledge can be distinguished in technical document interpretation: structural and syntactic knowledge, referring to the representation rules of objects, and semantic knowledge, referring to the objects themselves. It is shown how drawing analysis can be seen as a reading of the document in terms of discrete and non-discrete signs, the reading strategies being activated by specific symbols. An example of such a reading for a specific layer, annotations such as dimensioning, is presented.>

The authors try to explain why conversion of engineering drawings into CAD format has been stuck for so long in low-level operations such as vectorization, basic layer separation, and very limited symbol recognition, and what can be done to improve this situation. Based on their own expertise in syntactic and semantic analysis of technical drawings, they propose a scheme for achieving high-level conversion of engineering drawings into 3-D CAD models.>

It is shown that discrete relaxation can be used to provide a technical document with a consistent labeling. It can reduce the number of hypotheses associated with each graphical primitive of a drawing by exhibiting a locally consistent labeling of these primitives. The tools considered have been tested on cadastral city maps, and the results are discussed.>

A model-based approach is described for recognizing and locating polyhedral objects from sparse light-stripe data. Based on two scans, consistent interpretations are provided by locating scanned line segments on the faces of a set of known objects. The transformation associated with each interpretation is estimated. Once all the feasible models are transformed to the scene space, a strategy for additional scanning positions is developed to distinguish the object uniquely.>