0
$\begingroup$

I am working for a property portal and in our portal customers post several images with watermark and duplicity.

I have created a CNN that can detect good images and watermark, but I am unsure how can I train my model to detect duplicate images?

For example, if a customer posts 2 hall, 2 bedroom, 2 kitchen, how can I make my model to detect duplicate images so that I can take further action upon it like deletion from webpage.

Improve this question
$\endgroup$
4
  • $\begingroup$ When you say "duplicate" do you mean that the images are exactly the same (image A has exactly the same bytes as image B)? Or that they contain the same content (e.g. two pictures of the same kitchen, taken from slightly different angles)? $\endgroup$
    – Sycorax ♦
    Commented Jul 7, 2021 at 16:29
  • $\begingroup$ from different slightly angles $\endgroup$
    – James
    Commented Jul 7, 2021 at 17:45
  • 1
    $\begingroup$ triplet-loss might be helpful. FaceNet uses it to identify faces of the same person taken from different angles/different lighting. $\endgroup$
    – Sycorax ♦
    Commented Jul 7, 2021 at 18:13
  • $\begingroup$ I will try this definitely $\endgroup$
    – James
    Commented Jul 7, 2021 at 18:23

1 Answer 1

Reset to default
0
$\begingroup$

Wouldn't it be easier to use an Denoising autoencoder for this? and treat the watermark as gaussian noise that gets filtered in the hidden layer http://deeplearning.net/tutorial/dA.html

Improve this answer
$\endgroup$
1
  • $\begingroup$ I need to study it and will get back to you. Thank you so much for this, such a great help here $\endgroup$
    – James
    Commented Jan 30, 2018 at 4:52

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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