Handwritten synthetic dataset from the IAM

This dataset was generated employing a technique of randomly crossing out words from the IAM database, utilizing several types of strokes. The ratio of cross-out words to regular words in handwritten documents can vary greatly depending on the document and context. However, typically, the number of cross-out words is small compared with regular words. To ensure a realistic ratio of regular to cross-out words in our synthetic database, 30% of samples from the IAM training set were selected. First, the bounding box of each word in a line was detected. The bounding box covers the core area of the word. Then, at random, a word is crossed out within the core area. Each line contains a randomly struck-out word at a different position. The annotation of these struck-out words was replaced with the symbol #.

The folder has:
s-s0 images
Syn-trainset
Syn-validset
Syn_IAM_testset
The transcription files are in the format of
Filename, threshold label of handwritten line
s-s0-0,157 A # to stop Mr. Gaitskell from

Cite the below work if you have used this dataset:
"A deep learning approach to handwritten text recognition in the presence of struck-out text"
https://ieeexplore.ieee.org/document/8961024