ICFHR 2020 Competition on Image Retrieval for Historical Handwritten Fragments
Résumé
This competition succeeds upon a line of competitions for writer and style analysis of historical documentimages. In particular, we investigate the performance of large-scale retrieval of historical document fragments in terms ofstyle and writer identification. The analysis of historic fragmentsis a difficult challenge commonly solved by trained humanists.In comparison to previous competitions, we make the resultsmore meaningful by addressing the issue of sample granularityand moving from writer to page fragment retrieval. The twoapproaches, style and author identification, provide informationon what kind of information each method makes better use of andindirectly contribute to the interpretability of the participatingmethod. Therefore, we created a large dataset consisting of morethan 120 000 fragments. Although the most teams submittedmethods based on convolutional neural networks, the winningentry achieves an mAP below 40 %.