Comments on: Lookup tables for binary image processing—makelut and applylut

By: Steve

Steve — Mon, 07 Nov 2011 01:06:36 +0000

Tiago—A single 3-by-3 table operation isn’t enough to identify branch points, if that’s what you mean. But you can use the ‘branchpoints’ and ‘endpoints’ options of the bwmorph function.

By: Tiago

Tiago — Thu, 03 Nov 2011 01:09:29 +0000

Hello, I am not able to create a table to find ridge bifurcations and ridge ending. what would be the best table for this problem?

By: OJ

OJ — Sat, 09 Oct 2010 05:40:40 +0000

Steve, It worked like a charm. Thankyou.

By: Steve

Steve — Fri, 08 Oct 2010 11:59:07 +0000

OJ—Are you displaying the image at full magnification? When you display an image in a figure that’s too small to display all the pixels in the image, some pixels have to be discarded in the display, and that result in appearance of missing pixels. Trying using imtool and set the view magnification to 100%.

By: OJ

OJ — Fri, 08 Oct 2010 09:24:07 +0000

Hello Steve. I am new to Matlab Image Processing.

I am trying to thin image of a hand using ‘thin’. Apparently, bwmorph thinning produces image in which the pixels are missing at places. Somewhat like broken lines where they supposed to be continous.

By: Steve

Steve — Mon, 16 Jun 2008 13:40:18 +0000

Pete—I don’t think the lookup table optimization made in R2007b would make any difference for the operations you describe.

By: Pete

Pete — Mon, 16 Jun 2008 01:10:10 +0000

Interesting. I haven’t made much use of lookup tables. Although rather more limited, where possible I convert binary images to uint8 and use imfilter to add up the number of pixels in the neighborhood.

Therefore, for example, to compute end-points in a thinned/skeletonised binary image (bw) I have used,

bwEnds = imfilter(uint8(bw), ones(3)) == 2 & bw;

Crossover points might be,

bwCrossover = imfilter(uint8(bw), ones(3)) > 3) & bw;

My decidedly untrustworthy memory had told me that I compared both methods once upon a time and found imfilter to be faster, but when I tried them tonight (to locate end points in 768×768 pixel binary images) they each took almost exactly the same length of time (around 0.06 s). I wish to believe this is due to the R2007b optimisations rather than that I was mistaken, although I can’t be sure.

By: Steve

Steve — Tue, 03 Jun 2008 20:32:03 +0000

Mark—The lut method would probably be faster than what you are currently doing to detect end points. If a crossing configuration can be determined using only a 3-by-3 neighborhood, then that also could be done using applylut. Your suggestion about about adding these options to bwmorph is a good one; I'll follow up with you by e-mail about that.

By: Mark Hayworth

Mark Hayworth — Tue, 03 Jun 2008 17:36:02 +0000

Steve, it looks like this could be used to do two operations that are missing from bwmorph, and they are to identify crossings (triple and quad branch points) and (like you said) end points. Right now I’m getting end points by computing the spur and subtracting it from the original (is the lut method better than that method?). I have no way of computing crossings of skeleton lines where the shape is a “Y” or a “+” so maybe this could be used for that purpose. Even better would be to add these in to bwmorph as options – any chance of that?