ICFHR2016 Competition on Local Attribute Detection for Handwriting Recognition

This paper describes the ICFHR2016 Local Attribute Detection competition held at the 15th International Conference on Frontiers in Handwriting Recognition. This competition uses the ANDoc-Attrib-10K database, which consists of roughly 10,000 image snippets drawn from 30 collections from historical documents and records of various types, each labeled with seven attributes such as the presence of text, text type (handwritten machine-printed, or both), if the text is legible, etc. The objective of this competition is to use the training partition of the database to develop and submit for evaluation a system that detects the seven attributes in the snippets from the test partition. Two teams submitted five systems for scoring, evaluation, and comparison to Ancestry baseline systems. The results of this scoring are reported, along with brief descriptions of each system.