0

I am working on an image processing deep learning project using tensorflow. I calculated a brightness bias to add to all pixels of an image and calculated the LoG output of it. However, the sum of all pixel values was approximately the same as it was when I hadn't added the brightness bias.

It would have been drastically different if the LoG was linear in image pixel values since the brightness bias was very large. I just want to confirm that the LoG is supposed to stay the same so I can confirm that I'm not making a mistake elsewhere.

Harsh Motwani
  • 66
  • 7
  • 1
    It is going to stay the same in case the average of the Laplacian and Gaussian filer coefficients is 1 (and the border type is "symmetric", or constant with the save value of all the pixels). Can't you execute the LoG convolution on you input image and check? – Rotem Feb 18 '22 at 11:41
  • @Rotem I did. I got very similar values after adding a constant value and I got very similar results. That's why I wanted to make sure that I was executing it right. Thanks a lot for your response. – Harsh Motwani Feb 18 '22 at 18:18
  • @Rotem I think there's a correction. The average of coefficients should be zero because the constant value times the sum of all coefficients needs to be zero – Harsh Motwani Feb 18 '22 at 21:00
  • In case the filter is applies on the bias, the average needs to be zero. I referred to a filter that applies the input image. – Rotem Feb 18 '22 at 21:07

0 Answers0