Posted On: May 21, 2021

Amazon Rekognition is a machine learning powered image and video analysis service that can identify objects and concepts, people, faces, inappropriate content, as well as detect text. Rekognition text detection detects and reads text in an image, and returns bounding boxes for each word found. Starting today, Rekognition can detect up to 100 words in an image, up from the previous limit of 50 words. In addition, you get higher accuracy, especially for cases with illegible text, which are correctly rejected now. Lastly, the average latency for each text detection API call is reduced by up to 70%.

Text detection in images can be leveraged for multiple use cases. First, you can use text detection for content moderation and content redaction workflows, where detected text can be checked against your own list of inappropriate or unwanted words and phrases. Alternatively, you can use the detected text bounding box area to blur out sensitive text information. Second, text detection is useful in analysis and search of marketing assets. For example, you can understand how particular words, text placement, and text size impact campaign performance, or easily search for image or video assets with specific keywords or captions in your Digital Asset Management (DAM) system. Last, you can use text detection in mapping and automotive applications, to read text on road and street signs.

These improvements are now available to all customers at no additional cost. To get started, please visit the Rekognition Console or download the latest AWS SDK to start building applications. To learn more about text detection with Amazon Rekognition, see our documentation and FAQ.