Forgotten in its old age

by Aditya

submit your photo


Hall of Fame
View past winners from this year

Please participate in Meta
and help us grow.

Take the 2-minute tour ×
Photography Stack Exchange is a question and answer site for professional, enthusiast and amateur photographers. It's 100% free, no registration required.

I plan on taking a series of pictures of people in a room, keeping a fixed position of the camera, and later remove the background. Essentially, although the background is complicated, it should be approximately identical in all pictures.

Since this will be a large number of pictures I am looking for an (semi)-automated way of removing the background. I was thinking to some sort of background subtraction functionality such as Macs PhotoBooth, where, based on a reference background image, it automatically detect non background areas.

How is this possible on consumer applications such as Photoshop?

share|improve this question
3  
Not possible as far as I know. Even if the background is the same, the different sizes, shapes, and positions of the people will make automating removal hell. If the background was one solid colour like a green screen, it might be possible (assuming no one was wearing the colour), but with a complicated background? I don't think so. –  ElendilTheTall Nov 30 '13 at 22:40
    
How many pictures is a large number? Averaging the frames - stack them and then go from 100%, 50%, 25%, 12%, 6%, 3%, 2%, 1% opacity may get you close -- that said it is generally a pretty easy, but not automated job as well –  Patrick Hurley Nov 30 '13 at 23:52
3  
This task is easily accomplished in several ultra high end film editing apps, so it is certainly possible though I do not think possible with Photoshop. There may be some other app that can perform this function. I sure hope so –  Linuxmint Dec 1 '13 at 0:29

2 Answers 2

up vote 0 down vote accepted

Depending on how you make your photos, it can be pretty easy to impossible...

If

  • you control the lighting of the wall (it can be patterned, but makes life harder - in general a single color with even lighting is recommended)
  • there is no shadow being dropped by the people on the wall,
  • the wall color and ANY color on the people has no match (or they are not even near)
  • the camera is fixed
  • the DOF is fixed
  • the focus(!) is fixed
  • the white balance, etc. is fixed (use color calibration at the beginning)
  • and you use low ISO to avoid dots in the image,

then

  • you can take a reference photo of the wall,
  • make a photo of the people (or one person)
  • put these two images in Photoshop
  • use "Substract" on the two image layers
  • use threshold to select those points that are close to 0,
  • use the created image as a mask to mask the photo with the people

This all can be automated using batch processing. But before you do that, try it yourself manually. In general, this is not impossible, but not easy either.

You might need to mess a bit manually with masking, but if you keep to the rules above, the manual work is not so much. If you deviate from the rules above, it just gets harder and harder.

Especially around hair, this can become a nightmare, because hair works as a kind of filter, which can alter colors - this is just physics.

There is a reason why people use greenboxing/blueboxing, and even that needs good skills to produce great results (unless you go for a low-resolution output).

share|improve this answer

Computers only see pixels, not objects, so a background plate isn't going to be particularly helpful unless it is exactly the same. If it was exactly the same, a subtraction blend would do a pretty good job of removing it mostly, but minor exposure variation is still going to result in some noise.

Really though, there isn't any way to have it done automatically, extraction is a mostly manual process.

share|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

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