AWS Rekognition Text detection limited to 50 words

AWS Rekognition Text detection limited to 50 words

Feb 2022 - this post is permanently moved to https://architectfwd.com, my new site, and can be found here  - https://architectfwd.com/architecture/cloud/amazon-web-services-aws/rekognition/2022/01/23/aws-rekognition-text-detection-limited/ please go and bookmark that site for all of my future content.

I built an API which uses Rekognition on AWS quickly, just to test the text detection capability.

Approach 

I create a Flask API and utilised boto.

rek = boto3.client('rekognition', region_name="us-east-1")

After that I took the image bytes directly and ran a detect_text call, not too tough.

Detect text result

I uploaded an image with a small number of words and was pleased with the result. However when uploading an image containing a paragraph I found that only a subset of the words were returned.

The limit is 50 words - "DetectText can detect up to 50 words in an image."[0]

Text result response

The response splits up items by Type, either "line" or "word", and has a parentID when a word, so I filtered just the lines like this:

if label['Type']=="LINE"

It works, great result, but a solution for a larger number of words makes me think of just running this through Tesseract OCR.

[0] - https://docs.aws.amazon.com/rekognition/latest/dg/text-detection.html (Last paragraph)

Cheers

Quintes

Connect with me on LinkedIn or Twitter