Evaluating document analysis results via graph probing

While techniques for evaluating the performance of lower-level document analysis tasks such as optical character recognition have gained acceptance in the field, attempts to formalize the problem for higher-level algorithms that incorporate more complex structure have been less successful. We describe an intuitive, easy-to-implement scheme for the problem of performance evaluation when document recognition results are represented in the form of a directed acyclic graph. We present results from two simulation studies based on different graph models and one experiment using a well known page segmentation algorithm to demonstrate the applicability of the approach