Catálogo de publicaciones - libros

In the paper, we analyze the vectorization methods and errors of vectorization of monochrome images obtained by scanning line drawings. We focused our attention on widespread errors inherent in many commercial and academic universal vectorization systems. This error, an error of parity, depends on scanning resolution, thickness of line, and the type of vectorization method. The method of removal of parity errors is suggested. The problems of accuracy, required storage capacity, and admissible slowing of vectorization are discussed in the conclusion.

This paper presents a vectorization system for architecture engineering drawings. The system employs the line-symbol-text vectorization workflow to recognize graphic objects in the order of increasing characteristic complexity and progressively simplify the drawing image by removing recognized objects from it. Various recognition algorithms for basic graphic types have been developed and efficient interactive recognition methods are proposed as complements to automatic processing. Based on dimension recognition and analysis, the system reconstructs the literal dimension for vectorization results, which yields optimized vector data for CAD applications.

In this paper, we review some ideas which emerged in the early years of research on symbol recognition and we show how these ideas evolved into a large variety of contributions. We then propose some interesting challenges for symbol recognition research in the present years, including symbol spotting methods, recognition procedures for complex symbols, and a systematic approach to performance evaluation of symbol recognition methods.

In this paper we present a method to determine which symbols are probable to be found in technical drawings using vectorial signatures. These signatures are formulated in terms of geometric and structural constraints between segments, as parallelisms, straight angles, etc. After representing vectorized line drawings with attributed graphs, our approach works with a multi-scale representation of these graphs, retrieving the features that are expressive enough to create the signature. Since the proposed method integrates a distortion model, it can be used either with scanned and then vectorized drawings or with hand-drawn sketches.

In this paper, we present the problem of noisy images recognition and in particular the stage of primitives selection in a classification process. We suppose that segmentation and statistical features extraction on documentary images are realized. We describe precisely the use of concept lattice and compare it with a decision tree in a recognition process. From the experimental results, it appears that concept lattice is more adapted to the context of noisy images.

This paper describes a proposed extended system for the recognition and labeling of graphical objects within architectural and engineering documents that integrates Statistical Language Models (SLMs) with shape classifiers. Traditionally used for Natural Language Processing, SLMS have been successful in such fields as Speech Recognition and Information Retrieval. There exist similarities between natural language and technical graphical data that suggest that adapting SLMs for use with graphical data is a worthwhile approach. Statistical Graphical Language Models (SGLMs) are applied to graphical documents based on associations between different classes of shape in a drawing to automate the structuring and labeling of graphical data. The SGLMs are designed to be combined with other classifiers to improve their recognition performance. SGLMs perform best when the graphical domain being examined has an underlying semantic system, that is; graphical objects have not been placed randomly within the data. A system which combines a Shape Classifier with SGLMS is described.

In this paper, we investigates symbol representation introducing a new hybrid approach. Using a combination of statistical and structural descriptors, we overcome deficiencies of each method taken alone. Indeed, a Region Adjacency Graph of loops is associated with a graph of vectorial primitives. Thus, a loop is both representend in terms of its boundaries and its content. Some preliminary results are provided thanks to the evaluation protocol established for the GREC 2003 workshop. Experiments have shown that the existing system does not really suffer from errors but needs to take advantage of vectorial primitives which are not involved in the definition of loops.

The restoration and preservation of ancient documents is becoming an interesting application in document image analysis. This paper introduces a novel approach aimed at segmenting the graphical part in historical heritage called and extracting its signatures in order to develop a Content-Based Image Retrieval (CBIR) system. The research principle is established on the concept of invariant texture analysis (Co-occurrence and Run-length matrices, Autocorrelation function and Wold decomposition) and signature extraction (Mininum Spanning Tree and Pairwise Geometric Attributes). The experimental results are presented by highlighting difficulties related to the nature of strokes and textures in . The signatures extracted from segmented areas of interest are informative enough to gain a reliable CBIR system.

This paper describes a method of image processing for the 2D bar code image recognition, which is capable of processing images extremely rapidly and achieving high recognition rate. This method includes three steps. The first step is to find out the four vertexes of ROI (Regions Of Interest); the second is a geometric transform to form an upright image of ROI; the third is to restore a bilevel image of the upright image. This work is distinguished by a key contribution, which is used to find the four vertexes of ROI by using an integrated feature. The integrated feature is composed of simple rectangle features, which are selected by the AdaBoost algorithm. To calculate these simple rectangle features rapidly, the image representation called "Integral Image" is used.

We focus our attention on complex lithographic pattern generation on a huge substrate with no manipulation of the light source shape for Digital Micromirror Device (DMD) based maskless lithography. To overcome the limitations of existing pattern generation methods developed upon the assessment of lithographic paths of the reflected beam spots rather than the recognition of patterns, we place our primary concern on the pattern. We consider pattern generation for maskless lithography using the DMD as a graphic recognition field problem. The pattern generation process is conceptualized as dual pattern recognition in two contrary views, which are the substrate’s view and the DMD’s view. For pattern recognition in the DMD’s view, a unique criterion, the area ratio, is devised for approval of the on/off reflection of the DMD mirror. The Region-based Pattern Generation (RPG) scheme based upon the area ratio is proposed. For verification, a prototype RPG system is implemented, and lithography using the system is performed to fabricate an actual Flat Panel Display (FPD) glass. The results verify that the RPG scheme is robust enough to generate lithographic patterns in any possible lithographic configuration and the RPG system is precise enough to attain the lithographic quality required by the FPD manufacturer.