Multiresolution cooperation makes easier document structure recognition

Abstract

This paper shows the interest of imitating the perceptive vision to improve the recognition of the structure of ancient, noisy and low structured documents. The perceptive vision, that is used by human eye, consists in focusing attention on interesting elements after having detecting their presence in a global vision process. We propose a generic method in order to apply this concept to various problems and kinds of documents. Thus, we introduce the concept of cooperation between multiresolution visions into a generic method. The originality of this work is that the cooperation between resolutions is totally led by the knowledge dedicated to each kind of document. In this paper, we present this method on three kinds of documents: handwritten low structured mail documents, naturalization decree register that are archive noisy documents from the 19th century and Bangla script that requires a precise vision. This work is validated on 86,291 documents.

Converting heterogeneous statistical tables on the web to searchable databases

“…Similarities between table and form processing were already emphasized by Bing et al [25] and Kieninger and Dengel [26]. Grammar-based approaches that can be specialized to forms and tables have been demonstrated on large data sets [27,28,29]. However, bureaucratic forms often have preprinted labels rather than indexing headers like tables.…”

Athena

“…Exploiting a multiresolution representation and a frequency-based framework is an old and well-known approach in image analysis and segmentation (e.g., Sabharwal and Subramanya [2001]), which has been applied to a plethora of applications; generally speaking, it is used to treat in an adaptive way different kinds of input data (e.g., in terms of resolution and size of image details). In the specific field of document layout analysis, these methods are typically used to segment document images scanned from newspapers and journals [Qiao et al 2006;Lemaitre et al 2008]. Recently, Almeida and Almeida [2012] used wavelets to reduce ink show-through noise in scanned letters or images.…”

ATHENA: Automatic text height extraction for the analysis of old handwritten manuscripts

“…Exploiting a multi-resolution representation and a frequency-based framework is a wellknown approach in image analysis (e.g., [26]), which has been applied to a plethora of applications. In the specific field of document layout analysis these methods are typically used to segment document images scanned from newspapers and journals [27], [28]. Recently, Almeida et al [29] use wavelets to reduce ink show-through noise in scanned images.…”