Talk:Otsu's method

Javascript Variants
I really do not understand why there are two javascript implementations that are nearly identical. The only significant difference is that the second variant also provides code to load and convert an rgba image to grayscale and then apply otsu's method. The link to an offsite result page is weird, too. I suggest to remove the second variant. — Preceding unsigned comment added by 2A02:8071:308A:2100:7AE4:FF:FE4E:8683 (talk) 13:18, 18 May 2016 (UTC)

Untitled
In the first line it reads "... Otsu method ..." Shouldn't it be "Otsu's Method"? or "the Otsu method" also, the article is not found if you search for otsu's algorithm. ive seen it reffered to as such in a book and a few articles as both otsu's algorithm and otsu's method. someone should probably address this - Pikez33 Albert Cruz 03:29, 12 June 2008 (UTC)
 * Some references to articles that reffer to Otsu's method as Otsu's algorithm
 * Y. Liu and S. N. Srihari. Document Image Binarization Based on Texture Features. Pattern Analysis and Machine Intelligence, IEEE Transactions on. 19(5):540-544, 1997.
 * Y. Solihin and C.G. Leedham. Integral Ratio: A New Class of Global Thresholding Techniques for Handwriting Images. Pattern Analysis and Machine Intelligence, IEEE Transactions on. 21(8):761-768, 1999.
 * There are many more that I haven't included, but I think it's safe to say that it's also known as Otsu's algorithm. -Albert Cruz 18:18, 16 June 2008 (UTC) Pikez33
 * I went ahead and linked "otsus method" to this page, as it would be a common error when searching for this algorithm. will wait to link "algorithm" Albert Cruz 18:36, 16 June 2008 (UTC) Pikez33
 * done Albert Cruz 04:45, 30 June 2008 (UTC) Pikez33

Does anybody know if this method is known also as Ad-hoc algorithm. I do not think so, and the name seems weird to me, if nobody complains I will remove this. --Tomash (talk) 22:04, 16 February 2008 (UTC)
 * Removed. --Tomash (talk) 16:54, 1 April 2008 (UTC)
 * I did an extensive search and Otsu did not reffer to his image segmentation algorithm as ad-hoc. there are no references to an algorithm by Otsu under the name of ad-hoc algorithm/method in the context of image processing. Albert Cruz 18:18, 16 June 2008 (UTC) Pikez33

Visual example
this page needs a visual example of a grayscale image thresholded using Otsu's Albert Cruz 23:15, 1 July 2008 (UTC) Pikez33
 * I've linked a public domain image (original) and it's otsu's threshold result. they could probably be better formatted Albert Cruz 03:32, 3 July 2008 (UTC) Pikez33

General/Content
I'v changed from Category:Computer vision to Category:Computer vision stubs since this articles needs mora material to make it readable.--KYN 07:01, 22 March 2006 (UTC)

I thought this was a good description from the web - since the actual paper by Otsu is pre-internet: http://sampl.ece.ohio-state.edu/EE863/2003/ee863-9.ppt The thresholds are exhaustively searched (there are only 254 possible for greyscale images). For each threshold, Otsu aims to minimize variance*probability of points less than threshold + the variance*probability of the remaining points. This minimizes within-class variance, and maximizes between-class variance. In practice the variances are not actually computed, because a marginally more efficient incremental method was suggested by Otsu.
 * I changed the description so that it is clear that variances are not computed. --Tomash (talk) 22:10, 16 February 2008 (UTC)

I am currently a student at UC Riverside with some expertise in computer vision, and will try to improve this article - Albert Cruz 22:28, 8 August 2007 (UTC) Pikez33

Dynamic Range
The article states (misleadingly) that the threshold applies to the dynamic range of pixel intensities present in the image. This comment (and the reference provided for it) refers to the internal MATLAB function graythresh that implements Otsu's algorithm in this way. In fact the description of the algorithm given here, as well as all the pseudo-codes seem to be providing a real threshold value, and do not seem to be using the dynamic range. This is what it appears to me, but I have done neither experimentation, nor verification. If someone else does that, they could fix the article. If not, I will do so when I have to time to verify my claim. Arif Zaman (talk) 12:37, 29 October 2015 (UTC)

Now I have done a good amount of experimentation, and read the source code of MATLAB's implementation of Otsu's method. It is patently clear that the stackexchange article that was referred to was erroneous. I have made a comment with details there as well. Since most algorithms return a threshold value directly anyway, this comment is only with reference to a MATLAB function. I am going to remove the entire paragraph that creates this confusion. Arif Zaman (talk) 02:59, 30 October 2015 (UTC)

Psuedo-code
This pseudo code has nothing to do with Otsu's Method. This method's describes the iterative method to obtain the ideal threshold by Gonzalez and Woods [2002].

Pseudo code is an example of K-mean method... not Otsu's
 * Pseudo-code now reflects what I think is Otsu's algorithm. --Tomash (talk) 22:10, 16 February 2008 (UTC)

The code for the multi-level implementation using Matlab is not correct. The logic is not correct and there are programming errors. — Preceding unsigned comment added by 131.191.11.71 (talk) 18:20, 15 March 2016 (UTC)

Representation of Equations
this article seems to focus on otsu's method being used only in cases when there is a foreground and background. also, there is a mathematical representation of otsu's method

Category
The article states the method represents "histogram shape-based image thresholding". However it sounds more like a clustering-based method. —DIV (138.194.12.32 (talk) 08:40, 10 August 2010 (UTC))

Reversion of unsourced change
In, changed the number  .7152 to  .07152, apparently believing the number .7152 to have been the result of vandalism by an IP (see WP:HD.  He gives no evidence for the diagnosis of vandalism, and the IP's value of .7152 seems to be supported by a range of published sources, such as http://research.ijcaonline.org/volume49/number9/pxc3880757.pdf, https://rdrr.io/bioc/EBImage/man/channel.html, http://www.bioconductor.org/packages/release/bioc/manuals/EBImage/man/EBImage.pdf, and http://cmm.ensmp.fr/~marcoteg/cv/publi_pdf/amira/2016_IET_AdaptiveBinarisationSceneText.pdf (to name but a few).  I have therefore reverted Penubag's change, pending any further evidence. --David Biddulph (talk) 03:10, 8 May 2017 (UTC)
 * Run the code in your browser like it suggests. The correct output is only obtained by .07152, as was originally written in the first revision. See here -- penubag  (talk) 03:19, 8 May 2017 (UTC)


 * That sounds rather like WP:OR, but other users can judge that against the published sources. Do you have an example with the value .7152?  With further WP:OR it can be noted that in the variant , the 3 coefficients add up to 1, which may have been the intention;  experts in the field can perhaps confirm if necessary? --David Biddulph (talk) 03:29, 8 May 2017 (UTC)


 * FYI This is currently at ANI. In order to keep the discussion in one place, I suggest we keep it all there. Thanks. -Ad Orientem (talk) 03:31, 8 May 2017 (UTC)
 * Here is the image with .7152 which is incorrect. For now, I'll revert the IP's unsourced change because clearly it makes the example incorrect. -- penubag  (talk) 03:34, 8 May 2017 (UTC)


 * Those who understand such things may also wish to look at Grayscale (with the .7152 value). --David Biddulph (talk) 03:55, 8 May 2017 (UTC)


 * .7152 also occurs in Luma (video). --David Biddulph (talk) 03:58, 8 May 2017 (UTC)
 * Suggest that the most correct thing to do -- regardless of the correct green value -- would be to remove large amounts of OR (ie: code implementation) from the page.  ~Hydronium~Hydroxide~(Talk)~ 10:17, 8 May 2017 (UTC)