Talk:Connected-component labeling

Untitled

I intend to clean up this article and list a few of the common algorithms (2-pass rasterization for example). Charlie.liban (talk) 18:28, 11 February 2008 (UTC)

Done, and I completely forgot to add a description to the revision. Charlie.liban (talk) 05:13, 12 February 2008 (UTC)

Added an image for clarification. Charlie.liban (talk) 05:58, 12 February 2008 (UTC)

is the link to a spanish blog post really appropriate in the english article? —Preceding unsigned comment added by 146.6.202.194 (talk) 15:41, 5 August 2009 (UTC)

I don't think so. Removed. --Prydeson (talk) 21:37, 8 October 2009 (UTC)

Is it worth adding text about doing connected component labeling in a distributed parallel setting? There are significant scientific challenges to do doing this. SeanAhern (talk) 14:23, 6 October 2009 (UTC)

I would like to see a generalized algorithm for 3 or more dimensional labeling tinku99 —Preceding undated comment added 03:30, 23 November 2009 (UTC).

I'll look into it. Voxel CCL is of interest to biomedical imaging, so you may find some information in papers related to that. -charlie liban (talk) 17:06, 25 November 2009 (UTC)

It is quite hard to understand the pseudo code 'linked[label] = union(linked[label], L).. maby some explanations —Preceding unsigned comment added by 83.27.187.36 (talk) 12:56, 2 December 2009 (UTC)

The whole article is looking a bit cluttered for such a simple concept. I'll try and clarify this with better pseudocode. -charlie liban (talk) 21:02, 22 March 2011 (UTC)

This could still use some help. It's not clear what the linked sets are used for; they're never read, as near as I can tell. 76.104.197.185 (talk) 04:25, 8 March 2013 (UTC)

What language is the second source code in? —Preceding unsigned comment added by 117.192.203.26 (talk) 04:57, 17 May 2010 (UTC)

MATLAB from the looks of it -charlie liban (talk) 21:02, 22 March 2011 (UTC)

This is definitely not MATLAB: := is not a thing, and neither is end for. It could be Octave, but I doubt it. The end for would have to be endfor at the very least. I don't know if there's a := in Octave either. Mr. Neo Anderson (talk) 15:55, 5 May 2020 (UTC)

The pseudocode for the one-pass algorithm could be improved by linking to the union/find algorithm and by de-matlabing it (bsxfun). The goal should be that people can really understand how it works without running into "magic black box" functions. 87.149.159.77 (talk) 14:33, 11 March 2015 (UTC)

Maybe this algorithm should also be mentioned, it uses an entirely different approach, in one sequential scan all components are listed: http://telin.ugent.be/~jdebock/#cc — Preceding unsigned comment added by 77.109.78.111 (talk) 07:06, 15 July 2011 (UTC)

hi, what is the difference between region labeling and segmentation? thanks allot dowi84 — Preceding unsigned comment added by Dowi84 (talk • contribs) 16:48, 1 December 2012 (UTC)

Here is a possible source for "One component at a time": http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=4728561 --193.174.63.68 (talk) 13:11, 16 February 2015 (UTC)

Thank you, I added the "one component at a time" section. I was originally trying to implement the algorithms in this page, but they were very hard to implement and understand. Some time later, while I was reading Vincent and Soille's 1991 paper on Watershed segmentation, I saw that in one of their steps they had used this algorithm. It was fascinatingly easy to implement, fast and most importantly very easy to understand. So I added it here. I should have added the reference at that time. I am sorry for that. I will add it tomorrow. http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=87344&url=http%3A%2F%2Fieeexplore.ieee.org%2Fiel1%2F34%2F2845%2F00087344 Mohammad Akhlaghi (talk) 14:58, 16 February 2015 (UTC)

The "One-pass version" is (algorithmically) identical to the graph traversal (BFS or DFS). Deleting it would be a first step towards making this article better. 141.3.24.81 (talk) 15:37, 19 February 2016 (UTC)

Any particular reason for the "curlab" abbreviation? sounds like a brain fart — Preceding unsigned comment added by 46.59.161.153 (talk) 17:57, 29 February 2016 (UTC)

Don't need stack for 'One component at a time' DFS

Don't need stack for 'One component at a time' for depth-first search. Whatever is pushed on is immediately popped off! — Preceding unsigned comment added by Pdavidow (talk • contribs) 00:07, 16 September 2019 (UTC)

Graphic of connectivity

is a better graphic than the current 4-connected and 8-connected on this page. 206.223.175.102 (talk) 12:40, 8 May 2020 (UTC)

I believe the author of this graphic is trying to show that the CCL proceeds with preferred directions. What is missing it that pixel connectivity is limited to euclidean (up/down, left/right) vs allowing diagonals. I think that is too many concepts at once. I only guess the preferred direction from having seen an algorithm in ImageJ. For example Wand.java. 206.223.175.102 (talk) 12:47, 8 May 2020 (UTC)