6533b852fe1ef96bd12aacf7
RESEARCH PRODUCT
ICDAR 2021 Competition on Historical Document Classification
Dalia Rodríguez-salasAndreas MaierAnguelos NicolaouMathias SeuretVincent ChristleinDominique StutzmannNikolaus WeichselbaumerMartin Mayrsubject
Historical document imagesbusiness.industryComputer scienceDocument classificationDeep learningContrast (statistics)computer.software_genreVariety (linguistics)Task (project management)Competition (economics)Document classification[INFO.INFO-TS]Computer Science [cs]/Signal and Image ProcessingDocument analysisFontComputingMethodologies_DOCUMENTANDTEXTPROCESSINGDatingArtificial intelligence[SHS.HIST]Humanities and Social Sciences/HistorybusinesscomputerNatural language processingHistorical documentdescription
International audience; This competition investigated the performance of historical document classification. The analysis of historical documents is a difficult challenge commonly solved by trained humanists. We provided three different classification tasks, which can be solved individually or jointly: font group/script type, location, date. The document images are provided by several institutions and are taken from handwritten and printed books as well as from charters. In contrast to previous competitions, all participants relied upon Deep Learning based approaches. Nevertheless, we saw a great performance variety of the different submitted systems. The easiest task seemed to be font group recognition while the script type classification and location classification were more challenging. In the dating task, the best system achieved a mean absolute error of about 22 years.
| year | journal | country | edition | language |
|---|---|---|---|---|
| 2021-09-08 |