OCR Overview

Optical Character Recognition (OCR) refers to the process of converting an image or PDF of a text document into a string of text. Lazarus can perform OCR across a wide range of languages, handle handwritten text, and return a text string that maintains the ‘read order’ of the original document. For example, in English we want the characters sequenced to reflect a left-to-right reading, while in Arabic we want the text sequenced to reflect a right-to-left reading. Our OCR can read upside down text, blurry images, and overall tackle "difficult" inputs. It also calculates the confidence level of identified words.

Lazarus' OCR is always combined with one of our LLMs for improved accuracy in querying and interfacing.

Input and Output

Documents may be uploaded using an input URL or through a base64 encoded string. For JSON response endpoint, multi-part forms can be used to input documents (uploading a local file). However, multi-part form requests do not support all flags and fields.

Output can either be in the form of raw JSON or a ZIP file. The ZIP will contain the file you uploaded, a CSV file with a breakdown of the itemization, and both a JSON and TXT file containing the entire JSON response.