Textract doesn't recognize standar checkboxes

If you were a person instead of a chatbot, you would pass the query directly to the technical department, because the images have no problem. The image was generated by software so it is perfectly aligned and illuminated and has a resolution of 300DPI. The checkboxes are standard DevExpress report generator checkboxes. If you can advise me on the font that Textract recognizes best I would appreciate it. In the example you can see that the same text that appears several times in the form is sometimes recognized and sometimes not, as it happens with the checkboxes. Your model is advertised as one of the best, but it only works with the forms in your examples. In a real development the model is not applicable due to its inconsistency. My example is really simple and makes it clear that the model does not work. I have lost a week of work because it worked with simple examples and does not work with real world forms. Is there any way to improve the performance of the model in checkbox recognition?

I apologize, but it seems there's been a misunderstanding about Amazon Textract's capabilities with regard to checkbox detection. Amazon Textract actually does have the ability to detect selection elements, including checkboxes and radio buttons. This feature was added on April 24, 2019.

However, the accuracy of detection can vary depending on the quality and format of the input document. If you're experiencing inconsistent results, there are a few things you can try:

1. Ensure you're using the latest version of the Amazon Textract API. There have been several updates to improve accuracy, including enhancements to the forms extraction feature in June 2021.

2. Check the quality of your input image. Higher resolution, well-lit, and properly aligned images tend to yield better results.

3. Make sure you're using the appropriate API call. For form processing, you should use the AnalyzeDocument API with the "Forms" feature enabled.

4. If possible, try to standardize your form layout. Consistent formatting can help improve detection accuracy.

If you're still experiencing issues after trying these steps, you might want to consider using Amazon Textract in conjunction with Amazon Augmented AI (A2I). This allows you to implement human review for cases where the AI's confidence is low, which can help improve overall accuracy.

As for other AI models capable of detecting checkboxes, while there are other options available, Amazon Textract is generally considered one of the more capable and accurate services for this task. If you're having persistent issues, it might be worth reaching out to AWS support for more specific guidance tailored to your use case.
Sources
Document History for Amazon Textract - Amazon Textract
Amazon Textract announces updates to AnalyzeDocument - Forms
Compliance Validation for Amazon Textract - Amazon Textract