Create a model to read Name, expiry date on medicine image.
The images can be rotated, only portion of the name may be present, letters may be different in sizes etc. I have images data or I can generate more data if needed.
Please let me know your thoughts on how i can go about it.
Thanks,
You need to use NER named entity recognition, and I as far as I remember Tensorflow Developer or TensorFlow Advanced Techniques introduce it, but there are courses which introduce it at DLAI.
Check this link for more info:
The Natural Language Processing Specializations refers to this, more specifically Course 3 Week 2.
Wouldn’t a solution for this require isolating the sub region(s) of the image with item name and expiry date? I’m thinking license plate reader here…gotta find the plate before you ‘read’ it. If so, some variant of bounding box prediction. And then since the input is image, not text, the ‘reading’ might be more like NIST digit prediction than real NER operating on text. Thoughts?
If i understand properly that might the case too, although you would need a lot of images of different kinds of writtings to train the model I think. But it could surely work fine…once the model has been train properly.
I suggest this is conceptually less dissimilar than it might seem at first. Similar steps, and some of the same challenges for the OCR.
Thanks. Let me check and get back.
That is so coll man.