help with image processing (URGENT!!)
in [Matlab]
|
From: Ashish Uthama on 6 May 2010 16:18 On Wed, 05 May 2010 14:31:22 -0400, Jeremy Ho <jeremyho87(a)yahoo.co.uk> wrote: Do you have control on what gets places on the side of the spine? If you do: I would suggest using, say a hot pink colored paper :) This will make the image processing part _a lot_ simpler. If not, is the height from which the picture taken fixed? i.e will the scale of the paper with three black dots remain the same in all your pictures? If yes: consider using template matching. I tried a few approaches to detect circles in the image. While using EDGE, BWAREAOPEN and REGIONPROPS followed by some approach to detect 'circleness' of the regions, I did get close.. but the metal (?) eye holes in between the vertebrates also turn up as false positives. I cant see a way around it other than using the information of the orientation of the three holes in your paper (template based matching perhaps?)
From: Ashish Uthama on 6 May 2010 16:21 On Thu, 06 May 2010 16:15:21 -0400, Dan <samberpe(a)gmail.com> wrote: > > At the risk of turning a challenging problem into something mundane : Is > there any chance of taking color pictures? That would make identifying > the (now colored) dots trivial. > If not, you would have to make dots distinctive in some way.... which > *might* be tricky. > In any event, it would be helpful to have the actual image with the real > dots. > > Dan Chaining on this, even if the OP is limited to BW pictures, maybe he could use a more distinct shape instead of the dots. Maybe use a cookie cutter to cut out a shape :) or have a distinct pattern (even a simple checkerboard one would do I guess).
From: ImageAnalyst on 6 May 2010 16:26
On May 6, 4:18 pm, "Ashish Uthama" <first.l...(a)mathworks.com> wrote: > On Wed, 05 May 2010 14:31:22 -0400, Jeremy Ho <jeremyh...(a)yahoo.co.uk> > wrote: > > Do you have control on what gets places on the side of the spine? If you > do: I would suggest using, say a hot pink colored paper :) This will make > the image processing part _a lot_ simpler. > > If not, is the height from which the picture taken fixed? i.e will the > scale of the paper with three black dots remain the same in all your > pictures? If yes: consider using template matching. > > I tried a few approaches to detect circles in the image. While using EDGE, > BWAREAOPEN and REGIONPROPS followed by some approach to detect > 'circleness' of the regions, I did get close.. but the metal (?) eye holes > in between the vertebrates also turn up as false positives. I cant see a > way around it other than using the information of the orientation of the > three holes in your paper (template based matching perhaps?) ------------------------------------------------------------------------------------- I haven't had a chance to try it yet but it doesn't look too tough. Not trivial but not hard either. I'm certain that the white squares with dots can be considered to be just within some croppable area so that is one key - just crop out all the other stuff that complicates things. Then there are fewer things that you need to get rid of, which you can do just based on regular things like area, shape, intensity, EulerNumber (useful if each square is guaranteed to have at least one spot), etc. |