0
$\begingroup$

I have images that contain text and I am not sure what is the type of NN that can do the job for me. Basically, it will have to read the images and understand the text in order to be able to classify it. I am not sure if CNN is suitable in this case?

$\endgroup$
0
$\begingroup$

I would recommend using pretrained text extractors, like amazon textract or tesseract, and then train a network on the extracted text. The solution is going to be simpler and the learning is going to be easier for the NN.

| improve this answer | |
$\endgroup$
  • $\begingroup$ Actually I have also the original text (I convert it to image). I was just wondering if cnn are able to work on the text inside the images. Working with the text "as it is" is not helpful in my case $\endgroup$ – Lukas Jul 31 at 10:41
  • $\begingroup$ Yeah they can, and fully-connected networks can too, but why would you do that? I think Jeremy Howard from fast.ai mentions that in most cases is better for the network to be trained on two steps that end2end. I don't understand why working with text as is is not helpful in your case $\endgroup$ – David Masip Jul 31 at 10:45
  • $\begingroup$ So the answer is, CNN are not suitable, but nobody will prevent you from doing it $\endgroup$ – David Masip Jul 31 at 10:45
  • $\begingroup$ I think there are preprocessing steps that are better than converting to an image $\endgroup$ – David Masip Jul 31 at 10:57
  • $\begingroup$ Ok, thanks for your help $\endgroup$ – Lukas Jul 31 at 11:02

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.