Image Analysis OCR for Data and Content Compliance

In today's digital world, organizations have the challenge and responsibility of ensuring a safe and secure online environment for their users, employees, and partners. The increasing volume of images and videos shared on various social and communication channels necessitates robust data and content safety and compliance measures.

Optical Character Recognition (OCR) technology extracts text from images and scanned documents to make it machine-readable. This allows computers to read images' textual content and determine the location of the text within the image.

OCR in actionOCR in action


Content Compliance with Azure Image Analysis OCR

With the growing multimodal capabilities of large language models (LLMs), the extraction of textual insights from images is becoming increasingly essential. Organizations must ensure that the extracted image content is both harmless and compliant.

Text extraction from images with OCR facilitates identification of images containing harmful content such as profanity and hate speech. OCR-extracted text is passed to content moderation systems to classify and filter images with harmful text. The content moderation processes can be custom pipelines or leverage text and multimodal content moderation APIs offered by Azure AI Content Safety.

Optical Character Recognition (OCR) on an image with harmful textOptical Character Recognition (OCR) on an image with harmful text

Moderation strategies and policies should be tailored to align with the organization's unique goals and user needs. Some organizations use OCR to moderate content on images before the images are uploaded to LLM APIs such as GPT-4V turbo with vision. The text extracted from images is processed in-house moderation systems to get ratings on different safety categories. This prevents inappropriate and malicious text input from reaching the Large Language Model. This optimizes LLM API spending while also protecting LLMs from potentially malicious user activity.

Data Loss Prevention and Compliance with Image Analysis OCR

OCR can also be used to help identify and protect sensitive information in images. Private data such as health records, financial information, and Personally Identifiable Information (PII) like names, social security numbers, and addresses embedded in images can be detected with the help of OCR.

After OCR has extracted text from an image, the extracted text is passed to a sensitive data detector such as Azure AI PII Detection. The detector identifies and categorizes any sensitive information present in the text. This enables redaction or masking to prevent unauthorized access or sharing. Here's a code sample on how this can be accomplished with Azure OCR and PII Detection. 

Optical Character Recognition (OCR) on a picture of a National IDOptical Character Recognition (OCR) on a picture of a National ID

Utilizing OCR for the detection of sensitive data in images, which might otherwise have gone unnoticed, ensures adherence to privacy laws and industry standards for handling private information. This helps organizations build trust with users and customers while mitigating the risk of compliance violations.

High-Level Architecture DiagramHigh-Level Architecture Diagram

Example Customers

        Microsoft 365 logo.png       Microsoft News.png



Microsoft Purview Communication Compliance uses Azure Image Analysis OCR to extract text from images shared in Teams chat or Exchange online emails before running the text through a compliance pipeline. This process helps surface inappropriate content and sensitive information to compliance administrators for further action. Previously risky content embedded in images was not able to be detected and was a blind spot for compliance administrators.

Microsoft News also leverages Azure Image Analysis OCR to ensure that images embedded in news articles do not contain any inappropriate content.

Get Started with Azure AI Services Vision OCR

Moderate content and protect sensitive text information embedded in images using Azure AI OCR. Get started by following this code sample or our QuickStart guide. You can also use Vision Studio for a no-code try-out experience. Use Azure Document Intelligence if you are interested in OCR for documents. 



This article was originally published by Microsoft's Azure AI Services Blog. You can find the original article here.