0
$\begingroup$

Is it possible that using the Machine Learning classification model we can classify/predict the LOCATION/ORIENTATION of the lot number and expiration date on different vials label?

If yes, possible approaches.

Update: So far what I have observed is with EAST model https://arxiv.org/abs/1704.03155

one can locate text in the scene and then use OCR s/w like Textract to read those locations. Now, the real challenge is when you have lots of text available in the image vertical and horizontal and when you don't know where the particular text would appear in the image like expiry date - Is it possible to train a system to give just one or two locations which will be expiry date and lot number?

The challenge here I feel is the location but what features we would take to train as it's just text. I feel it's a classification problem with supervised learning but I may be wrong. Comment?

Thanks

$\endgroup$
  • $\begingroup$ Tell me this is not another assignment dumped on an unsuspecting community. $\endgroup$ – greybeard Apr 20 at 9:30
  • $\begingroup$ Nope, this is a genuine problem I am trying to solve. I have researched EAST model and Textract and other research papers. I know what I am asking.. $\endgroup$ – SmartObject Apr 20 at 15:24
  • $\begingroup$ Consolation, at last! (What made me uneasy was not asking for the most promising approach, mentioning the above; but restricting the question to it and requesting a reasoning when suggested infeasible (we can not).) $\endgroup$ – greybeard Apr 20 at 15:26
  • $\begingroup$ Removed reasoning if that makes easy to answer $\endgroup$ – SmartObject Apr 21 at 15:08
  • $\begingroup$ It would help to tell us what approaches you've already researched and considered. Provide a little background and make this useful and interesting for others. $\endgroup$ – D.W. Apr 22 at 4:52
0
$\begingroup$

It seems plausible that it might be doable, using methods in the research literature on object detection and localization (Yolo, Faster R-CNN, etc.). The only way to tell for sure is to acquire a large dataset of such images and try it out and see how well it works on your data.

| cite | improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.