Catálogo de publicaciones - libros

Text detection and tracking is an important step in a video content analysis system as it brings important semantic clues which is a vital supplemental source of index information. While there has been a significant amount of research done on video text detection and tracking, there are very few works on performance evaluation of such systems. Evaluations of this nature have not been attempted because of the extensive effort required to establish a reliable ground truth even for a moderate video dataset. However, such ventures are gaining importance now. In this paper, we propose a generic method for evaluation of object detection and tracking systems in video domains where ground truth objects can be bounded by simple geometric shapes (polygons, ellipses). Two comprehensive measures, one each for detection and tracking, are proposed and substantiated to capture different aspects of the task in a single score. We choose text detection and tracking tasks to show the effectiveness of our evaluation framework. Results are presented from evaluations of existing algorithms using real world data and the metrics are shown to be effective in measuring the total accuracy of these detection and tracking algorithms.

When a user places a document in a capture device—copier, multi-functional printer [MFP], or scanner—the user expects good output to be produced regardless of the document type. There are a variety of means to achieve improved output, in which the settings on the copying device are tuned to the content characteristics of the document. These settings can be automated across the range of scanned context extremes from photo (blurring, no snapping) to fully-text (sharpening, aggressive snapping) documents. This procedure is “document auto typing”, and relies on a fast and accurate assessment of the content of the captured image. We herein describe the development of seven distinct systems for document analysis, and through the comparison of these systems arrive at an efficient and accurate document analysis system for automating the copying settings. We discuss the applicability of this method to other automated workflows in document capture.

There are numerous cadastral maps generated by past land surveying. The raster digitization of these paper maps is in progress. For effective and efficient use of these maps, we have to assemble the set of maps to make them superimposable on other geographic information in a Geographic Information System. The problem can be seen as a complex jigsaw puzzle where the pieces are the cadastre sections extracted from the map. We present an automatic solution to this geographic jigsaw puzzle, based on the generalized Hough transformation that detects the longest common boundary between every piece and its neighbors. The experiments have been conducted using the map of Mie Prefecture, Japan and the French cadastral map. The results of the experiment with the French cadastral maps show that the proposed method, which consists of extracting an external area and extracting and regularizing the north arrow, is suitable for assembling the cadastral map. The final goal of the process is to integrate every piece of the puzzle into a national geographic reference frame and database.

This paper presents some first steps in building an interactive system which allows a user to efficiently browse a large set of scanned documents, without prior knowledge on the content of these documents, and retrieving symbols of interest to him personally, through a relevance feedback mechanism.

In this paper we present the algorithm of our fuzzy-spatial descriptor for symbol recognition in on-line sketches. Fuzzy sets are used to enhance this approach to cope with the inherent distortion of freehand drawings. It allows handling some primitive irregularities due to different ways of drawing in complex composite forms. The descriptor provides complete local information of the primitives that consist of only straight lines. This descriptor used the opening angle between primitives and its overlapping properties in 1D projection to describe its local spatial properties. Segmentation by curve and pen-speed properties of each drawing has been used in obtaining our basic primitives. The experiment results show that a part from its rotational-invariant, this descriptor is also robust to multi-scale of segmentation of its primitives.