ABBYY has integrated AI into its new text recognition system, which allows it to independently determine the peculiarities of handwriting or printing on individual images.
Although programs for translating scans into an editable file appeared in the last century, their main problem was recognition by comparing elements a snapshot with a set of symbols in their database. Therefore, when processing images with minor changes, often the wrong letters were displayed, which required proofreading and manual correction of errors.
Russian developer ABBYY solved this problem using neural networks. The presence of AI elements in the recognition system allows it to work like a human brain, determining the individual characteristics of the spelling of letters. At the same time, it does not wait for the database to be updated, but independently compares the sets of scanned characters with those already known, and expands the list of possible spellings. Details can be found in patent.
This speeds up the processing of scans with imperfections, documents with displaced lines, poorly printed elements, and images with hieroglyphs, Arabic script.
However, competition in this market continues to intensify. In May, Amazon also introduced a new Textract service AI-powered document processing that fully automates data entry.
text: Ilya Bauer, photo: kommersant