Modify Analyzed Document Field Bounding Boxes

Question

Modify Analyzed Document Field Bounding Boxes

CHIPPENDALE,TOM (Agilent USA) 21

I am using the Form Recognizer 3.0 version and I was wondering if there is a way to modify the bounding box that is specified for each field after analyzing the document. Is there a way manually edit the bound box selected by the layout analyzer. For instance, in the attached image only the "2" is highlighted in the last column. I would like to expand the selection so that the "<2" is highlighted. I suppose you could modify the JSON files that are generated by the tool, but it seems like there should be a way to make changes from the user interface. Is there a way to perform field bounding box modifications from the user interface?

romungi-MSFT 48,906 Reputation points Microsoft Employee Moderator

2021-12-13T11:30:59.477+00:00

@CHIPPENDALE,TOM (Agilent USA) The current version of the studio or sample labeling tool does not provide the functionality to edit the bounding box that is recognized while loading the document in the tool. You can create tags and select the text but that is not the same as updating the bounding box after OCR is run on the document. I think in your case if the document quality was better the character < would have been recognized. Do you have another document or image with a better quality to test the same?
CHIPPENDALE,TOM (Agilent USA) 21 Reputation points

2021-12-13T15:30:23.767+00:00

The bounding box that is generated by the form recognizer analyzer is correct most of the time (i.e. > 80%) . However, I have found some images that are incorrectly annotated. Unfortunately, I do not have complete control over the quality of the document images that we are using for the application. There is always going to be some variation in the document quality. I was hoping that there was a way to inform the image analyzer that the bounding box chosen was incorrect so that the performance of the training could be improved.

Your answer

romungi-MSFT 48,906 Reputation points Microsoft Employee Moderator

2021-12-13T11:30:59.477+00:00

@CHIPPENDALE,TOM (Agilent USA) The current version of the studio or sample labeling tool does not provide the functionality to edit the bounding box that is recognized while loading the document in the tool. You can create tags and select the text but that is not the same as updating the bounding box after OCR is run on the document. I think in your case if the document quality was better the character < would have been recognized. Do you have another document or image with a better quality to test the same?
CHIPPENDALE,TOM (Agilent USA) 21 Reputation points

2021-12-13T15:30:23.767+00:00

The bounding box that is generated by the form recognizer analyzer is correct most of the time (i.e. > 80%) . However, I have found some images that are incorrectly annotated. Unfortunately, I do not have complete control over the quality of the document images that we are using for the application. There is always going to be some variation in the document quality. I was hoping that there was a way to inform the image analyzer that the bounding box chosen was incorrect so that the performance of the training could be improved.

Share via

Modify Analyzed Document Field Bounding Boxes

Your answer