Learning nongenerative grammatical models for document analysis

Abstract

We present a general approach for the hierarchical segmentation and labeling of document layout structures. This approach models document layout as a grammar and performs a global search for the optimal parse based on a grammatical cost function. Our contribution is to utilize machine learning to discriminatively select features and set all parameters in the parsing process. Therefore, and unlike many other approaches for layout analysis, ours can easily adapt itself to a variety of document analysis problems. One need only specify the page grammar and provide a set of correctly labeled pages. We apply this technique to two document image analysis tasks: page layout structure extraction and mathematical expression interpretation. Experiments demonstrate that the learned grammars can be used to extract the document structure in 57 files from the UWIII document image database. We also show that the same framework can be used to automatically interpret printed mathematical expressions so as to recreate the original LaTeX.

Force deployment analysis with generalized grammar

“…It is already known that some of the complexity results for levels of the Chomsky hierarchy do not carry over [75,62], and a careful mathematical exploration of the area is much needed.…”

“…In multiset grammars mðÁÞ 0. Other topologies can be specified; for example, in document structure analysis, twodimensional topologies are required [59][60][61][62]. Symbols may be tuples; for example, in most natural language applications, features such as number and gender may be added to the basic symbols.…”

Structural Mixtures for Statistical Layout Analysis

“…Most of the efforts made by other researchers towards the development of trainable layout analysis systems have focused on the use of probabilistic grammars [10,18,20]. The latest development in the domain of grammatical modeling of document layouts is by Shilman et al [18].…”

On the Reading of Tables of Contents

“…First, algorithms will have to be capable of gathering, and merging evidence and decisions at multiple scales of granularity. Second, functional segmentation or grouping of visual elements is exponentially complex [11]. The grouping of pixels (or characters if the TOC is in PDF format) into semantically valid chunks, their functional role interpretation, and reconciliation of individual functional roles to the overall syntactic and semantic structure will have to proceed hand in hand, perhaps through iterative steps of evidence exchange.…”