The vast majority of algorithms in the table recognition literature are specified informally as a sequence of operations [7]. This has the undesirable side effects that models of table structure are implicit, defined generatively by the sequence of operations, and that the effects of intermediate decisions are often lost as usually a single interpretation is modified in-place.

We wished to compare the Handley [2] and Hu et al. [4]. table structure recognition algorithms and the complete set of table cell hypotheses they each generated, including any rejected in the final result. Rebuilding the systems using procedural code that transformed data structures for interpretations in-place would not have achieved this goal. Initially we translated the strategies to a formal model-based (specifically grammarbased) framework. A well designed model-driven system (such as DMOS by Couasnon ¨ [1]) makes it easier to observe and record decision making, and can be programmed succinctly by a model specification. However, we found mapping the sequence of operations in the strategies to a model based description was difficult, and the formal system required frequent and substantial reconfiguration in order to incorporate unanticipated requirements.

We then considered an intermediate level of formalization. By using a small set of basic graph-based operations we could define recognition algorithms as a series of decisions, where the alternatives for each decision were model operations of a specified type (e.g. classifying table cells as header cells or data cells). This made the model operations considered and applied at each decision point explicit, permitted dependencies between logical types to be automatically recovered, and allowed the complete history of hypothesis creation, rejection, and reinstatement to be automatically captured. The resulting formalization is the Recognition Strategy Language (RSL).

Date of creation, presentation, or exhibit



©2006 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE.

Document Type

Conference Proceeding

Department, Program, or Center

Computer Science (GCCIS)


RIT – Main Campus