A comprehensive survey of handwritten document benchmarks: structure, usage and evaluation

EURASIP Journal on Image and Video Processing

Table 4 Overview of state-of-the-art results on commonly used databases

Database	Task	Experimental protocol	Evaluation metric	Best results
IFN/ENIT	Handwriting recognition	Train: 32,492 words, Test: 8671 words	Recognition rate	93.37 % [217]
	Writer identification	411 writers	Top 1 Identification Rate	90 % [219]
QUWI	Writer identification	206 Writers, 2 paragraphs in training 1 in test	Top 1 Identification Rate	95.30 % [149]
	Gender Classification	Train: 282 writers, test: 193 writers	Classification rate	69.25 % [150]
	Multi-script writer identification	Train: 300 writers, test: 100 writers	Top 1 Identification Rate	55 % [220]
	Multi-script gender classification	Train: 300 samples, test: 100 samples	Classification rate	65 % [220]
IAM	Handwriting recognition	Varied	Recognition rate	80-90 % [215, 216]
	Writer identification	657 writers, one sample each in training & test	Top 1 Identification Rate	96.7 % [218]
RIMES	Word recognition	Training: 50,000 words, validation: 7000, test: 5740 words	Recognition rate	94.87 % [69]
KHATT	Writer identification	1000 writers with 4 samples/writers in training, test: 1822 images	Top 1 Identification Rate	73.4 % [231]
CVL	Digit recognition	Train: 7,000, validation: 7,000, test: 21,780	Precision	97.74 % [115]
CASIA	Character recognition	3755 character classes	Recognition rate	Offline:94.77 %, online: 97.39 % [229]