-2

I was trying hard to find a solution for automated background remove. Target images are modeling images with white/gray gradient background which are sometimes a nightmare for background remove. For example I tried this image which the model wears white dress and background is also white/gray:

before

after

As you can see, there is much noise in the image, and it's not because of the quality. It's because of using gimp>color>levels. I use it to make image edges stronger so edge detector can detect it.

and if you are interested, this is how I achieved the results:

colors > level -> 0- 0.1 - 255
filters> blure > blure (default) **optional**
filters > edge detector > edge... >smear 2.0
colors> hue:180  lightness/saturation:-100

How can I remove this noise (automatically, not manually)? Do you have any better idea for auto background remove from these kind of images?

Thanks in advance.

Hack-R
  • 22,422
  • 14
  • 75
  • 131
imaraz
  • 11
  • 1
  • I have edited your question. In the future please feel free to use proper capitalization and punctuation. – Hack-R Jul 30 '16 at 18:54
  • Please explain why you just tried to rollback corrections to your capitalization, grammar and punctuation? Vandalism, even of your own question, may result in your account not being allowed to ask or edit questions. – Hack-R Jul 30 '16 at 18:56
  • hi, i think that was a coincidence, i edited text right after posting it. you did edit it on same time frame i guess. sorry for that ;) – imaraz Jul 31 '16 at 11:38

1 Answers1

1

Doing some strong processing on the image to detect edges or else doesn't mean that you need to use this image afterwards. You can copy the image layer, detect edges on it, and apply the final selection on the original image layer.

Also, when you use any kind of global color processing you lose colors (use Color>Info>Color cube analysis to see how bad it can get). This goes worse with every color editing step, and pixels that originally have very close levels can become separated by a more visible level gap, which shows up as noise or banding. Using one single color transform (usually with Curves, since it is the most versatile) mitigates the problem. You can also try to use the current development version (2.9), that can use 16- or 32-bit channel depths which makes color loss much less of a problem.

xenoid
  • 8,396
  • 3
  • 23
  • 49