Using DetectText in rekognition to extract text from images of low quality.

Question

I am trying to extract text from an image using DetectText. Which works fine at least to some level. I want to be able to extract text from a small region of the image. which being small naturally makes the text small in character. And given the fact that the quality of the image is quite small. the accuracy is further reduced.

For reference on which regions of the image am talking about. Have a look at the attached image. I have highlighted the regions with blue and orange circles. Enter image description here

For reference on which regions of the image am talking about. Have a look at the attached image. I have highlighted the regions with blue and orange circles.![Enter image description here](https://repost.aws/media/postImages/original/IMQAe2JQLyS-S4LktyZsUWmA)

Answer

I'd say your best bet is to focus on upscaling the resolution of the small piece before handing it over to TextTract. Using libraries such as https://www.npmjs.com/package/upscaler should allow you to enhance the image and then https://www.npmjs.com/package/image-brightness should raise the image quality to try give it better accuracy.

Using DetectText in rekognition to extract text from images of low quality.

関連するコンテンツ