This paper presents an overview of feature extraction methods for off-line recognition of segmented (isolated) characters. Selection of a feature extraction method is probably the single most important factor in achieving high recognition performance in character recognition systems. Different feature extraction methods are designed for different representations of the characters, such as solid binary characters, character contours, skeletons (thinned characters), or gray level subimages of each individual character. The feature extraction methods are discussed in terms of invariance properties, reconstructability, and expected distortions and variability of the characters. The problem of choosing the appropriate feature extraction method for a given application is also discussed. When a few promising feature extraction methods have been identified, they need to be evaluated experimentally to find the best method for the given application. Feature extraction Optical character recogniti...

future trends

We present a new method for the skeletonization of 2-dimensional or 3-dimensional objects. First, we introduce two local measures, φ and d, to characterize skeleton points, whose good localization is ensured by Euclidean distance mapping techniques. These measures allow us to control the detail of the resulting skeleton. Thresholding these measures generally does not yield well-defined skeleton: a low threshold preserves the original object's topology but produces a noise-sensitive skeleton, while a higher threshold produces a more robust skeleton but it is generally not homotopic with the original object it comes from. To overcome these drawbacks, more complex measures can be introduced. Although they generally yield good experimental results, they are still sensitive to noise. Instead, we introduce a global step, called topological reconstruction, which will provide the skeleton with robustness with respect to noise and ensure homotopy with the original object. Moreover, this method is not...

Abstract. This paper presents a novel processing scheme for the automatic and robust computation of a medial shape model, which represents an object population with shape variability. The sensitivity of medial descriptions to object variations and small boundary perturbations are fundamental problems of any skeletonization technique. These problems are approached with the computation of a model with common medial branching topology and grid sampling. This model is then used for a medial shape description of individual objects via a constrained model fit. The process starts from parametric 3D boundary representations with existing point-to-point homology between objects. The Voronoi skeleton of each sampled object boundary is partitioned into non-branching medial sheets and simplified by a novel pruning algorithm using a volumetric contribution criterion. Using the surface homology, medial sheets are combined to form a common medial branching topology. Finally, the medial sheets are sampled and represented as meshes of medial primitives. Results on populations of up to 184 biological objects clearly demonstrate that the common medial branching topology can be described by a small number of medial sheets and that even a coarse sampling leads to a close approximation of individual objects.

We present a new method to reconstruct an implicit representation of a branching object from a set of data points scattered on its surface. The method is based on the computation of a geometric skeleton inside the data set. This skeleton is simplified in order to filter noise and converted into skeletal elements -- a graph of interconnected curves -- that generate an implicit surface. We use Bezier triangles as extra skeletal elements to perform bulge free blends between branches while controlling the blend extent. The result is a smooth reconstruction of the object, that can be computed whatever its topology. The skeleton o#ers compact storage, and provides an underlying structure for the reconstructed object, making it easier to edit in a modeling or animation environment. 1. Introduction With the recent developement of advanced rangeimaging sensors, automatic reconstruction of realworld objects from scattered data points has become an important issue in Computer Graphics. Applicati...

For liver surgical planning, the structure and morphology of the hepatic vessels and their relationship to tumors are of major interest. To achieve a fast and robust assistance with optimal quantitative and visual information, we present methods for a geometrical and structural analysis of vessel systems. Starting from the raw image data a sequence of image processing steps has to be carried out until a 3D representation of the relevant anatomic and pathologic structures is generated. Based on aT scans the following steps are performed: (1) The volume data is preprocessed and the vessels are segmented. (2) The skeleton of the ves- sels is determined and transformed into a graph enabling a geometrical and structural shape analysis. Using this information the different intrahepatic vessel systems are iden- tified automatically. (3) Based on the structural analysis of the branches of the portal vein their vascular territories are approximated with different methods. These methods are compared and validated anatomically by means of cor- rosion casts of human livers. (4) Vessels are visualized with graphics primitives fitted to the skeleton to provide smooth visualizations without aliasing artifacts. The image analysis techniques have been evaluated in the clinical environment and have been used in more than 170 cases so far to plan interventions and transplantations.

Homotopic thinning algorithms have long been investigated in pattern recognition and image analysis. However, they are order dependent in the sense that the output depends upon the order used for processing the image pixels (as well as the order in which homotopic structuring elements are applied for algorithms based on sequential homotopic thinning). In this paper, we tackle this problem by introducing the notion of order independent homotopic thinning. In its basic form, it leads to an order independent homotopic marking algorithm. However, when considering appropriate anchor points, it leads to binary and grey tone order independent skeletonisation algorithms. An application to the extraction of crest lines on digital elevation models is briefly presented before concluding.

Iso-surfaces are routinely used for the visualization of volumetric structures. Further processing (such as quantitative analysis, morphometric measurements, shape description) requires volume representations. The skeleton representation matches these requirements by providing a concise description of the object. This paper has two parts. First, we exhibit an algorithm which locally builds an iso-surface with two significant properties: it is a 2-manifold and the surface is a subcomplex of the Delaunay tetrahedrization of its vertices. Secondly, because of the latter property, the skeleton can in turn be computed from the dual of the Delaunay tetrahedrization of the iso-surface vertices. The skeleton representation, although informative, is very sensitive to noise. This is why we associate a graph to each skeleton for two purposes: (i) the amount of noise can be identified and quantified on the graph and (ii) the selection of the graph subpart that does not correspond to noise induces a filtering on the skeleton. Finally, we show some results on synthetic and medical images. An application, measuring the thickness of objects (heart ventricles, bone samples) is also presented.

Many classical image processing tasks can be realized as evaluations of a boolean function over subsets of an image. For instance, the simplicity test used in 3D thinning requires examining the 26 neighbors of each voxel and computing a single boolean function of these inputs. In this article, we show how Binary Decision Diagrams can be used to produce automatically very efficient and compact code for such functions. The total number of operations performed by a generated function is at most one test and one branching for each input value (e.g., in the case of 3D thinning, 26 tests and branchings). At each stage, the function is guaranteed to examine only the pertinent input data, i.e., the values which affect the result. As an example, we consider the 2D and 3D simplicity tests in digital topology, and thinning processes. We produce functions much faster than our previously optimized implementations [18, 4], and than any other implementation we know of. In the case of 3D simplicit...

Recognition of Handwritten Numerals Using Elastic Matching Patrice Scattolin Elastic matching has been used for the recognition of handwritten characters for two decades. It is usually only used for writer-dependent systems with on-line data. We attempt to use this method in a multi-writer environment for both on-line and off-line recognition of handwritten numerals. By its nature, elastic matching is best suited to single writer on-line systems. Two challenges present themselves to attain reasonable results under these conditions. First, the algorithm must be modified to better generalize the models, to recognize a wider variety of patterns with a given number of models. Secondly the off-line data is not in a suitable format as the patterns are not represented by a sequence of ordered points. We will apply two modifications to the typical elastic matching system to adapt it to the multi-writer environment and for the off-line data. To process the off-line data, we use a stroke recon...