Scanned paperwork is searchable only once it has text. Fieldnote reads each page with OCR once, when it is added, keeps every line with its position, and answers questions like “the total on the Popular Book receipt from 5 March” by finding the page, reading the value, and showing the crop of the page it came from, so a person can check it at a glance. Below: questions over 200 scanned receipts.
The question names the shop and one known fact; the index has to find the right page among 200 and read the other. The crop is the citation: the row the answer was read from.
Loading…
The cited crop
200 receipts, 400 questions
The same index built twice: on text read here with OCR, and on the dataset's hand transcription, which shows how much the OCR costs.