A platform for storing, visualizing, and interpreting collections of noisy documents

Abstract

The goal of document image analysis is to produce interpretations that match those of a fluent and knowledgeable human when viewing the same input. Because computer vision techniques are not perfect, the text that results when processing scanned pages is frequently noisy. Building on previous work, we propose a new paradigm for handling the inevitable incomplete, partial, erroneous, or slightly orthogonal interpretations that commonly arise in document datasets. Starting from the observation that interpretations are dependent on application context or user viewpoint, we describe a platform now under development that is capable of managing multiple interpretations for a document and offers an unprecedented level of interaction so that users can freely build upon, extend, or correct existing interpretations. In this way, the system supports the creation of a continuously expanding and improving document analysis repository which can be used to support research in the field.

Document Analysis Research in the Year 2021

“…In this paper, we have presented a vision for the future of experimental research in document analysis and described how our current DAE platform [8,9] can exploit collective intelligence to instill new practices in the field. This forms a significant first step toward a crowd-sourced document resource platform that can contribute in many ways to more reproducible and sustainable machine perception research.…”

“…We also refer to a proof-of-concept prototype platform for Document Analysis and Exploitation (DAE -not to be confused with DARE), accessible at http://dae.cse.lehigh.edu, which is capable of storing data, meta-data and interpretations, interaction software, and complete provenance as more fully described elsewhere [8,9]. DAE is an important step in the direction of DARE, but still short of the grand vision described earlier.…”

“…Annotations need not be human-contributed but can be the result of complete document analysis pipelines and algorithms. As a result, the resource can hold apparently contradictory interpretations of identical documents, when these stem from different transformation and analysis processes [8].…”

The Non-geek's Guide to the DAE Platform

“…As already described in [4], the DAE platform stores more than just basic collections of documents. It supports a data model that, from a higher level viewpoint, incorporates concepts and relations summarized by the following set of axioms:…”

“…Previous accounts of the features and the potential uses of the platform [4], [5], [7] focused on its architecture, implementation choices, and its potential to impact experimental research, especially with respect to reproducibility and accountability.…”