Skip to main content
R
Richard279255950ecc
Participant
September 11, 2023
Question

Best way to select a standard object in an image

  • September 11, 2023
  • 1 reply
  • 528 views

Hi!

I have an automation that removes the background of a photo of a framed comic, straightens the image and then resizes to a standard size. This generally works ok - see below before and after.

 

On a few of the photos the selection doesn't always sucessfully select the frame outline depending on the shadows/lighting. I can fix this by tweaking the tolerance and sampling but then it then works on some and not the others. I have tried "subject select" and "removed background" but neither of these works reliably.

 

Is there a more robust way to select the frame e.g. using the fact it's always a simple 4 sided polygon

 

Thanks!

 

 

 

    This topic has been closed for replies.

    1 reply

    c.pfaffenbichler
    Community Expert
    c.pfaffenbichlerCommunity Expert
    Community Expert
    September 12, 2023
    quote

    Is there a more robust way to select the frame e.g. using the fact it's always a simple 4 sided polygon

    It is commendable that you posted meaningful images right away, but that question seems a bit optimistic. 

    How would you define a »four sided polygon« in a pixel photograph exactly? (Due to the lenses the »lines« connecting the corners aren’t even straight …) 

     

    Have you tried involving the Filter »Find Edges«? If the background is always fairly unform this might provide a good basis for differentiating the foreground. 

    R
    Richard279255950eccAuthor
    Participant
    September 12, 2023

    Hi! Many thanks and the "Find Edges" tip is great and has got me closer! I can select the non-picture area now a lot cleaner and I'm just working through how to then use this as a mask for the original image - almost there but need more googling!

      c.pfaffenbichler
      Community Expert
      c.pfaffenbichlerCommunity Expert
      Community Expert
      September 12, 2023

      Personally I often like a Work Path approach, because one can get all the points of the path and process them with a Script.

      But identifying a rectangle that way seems difficult because a Work Path based on a Selection of the (pre-processed) image will still likely contain dozens (or hundreds or …) of points, so identifying the »exact« corners would probably need employing some heuristics. 

      Terms & ConditionsAccessibility statement
      Using the community Terms of use Privacy Cookie preferences Do not sell or share my personal information ad choicesAdChoices