Fred's ImageMagick Scripts



 

 

OTSUTHRESH


Automatically thresholds an image to binary (b/w) format using Otsu's between class variance technique.

Download Script

last modified: October 29, 2008



USAGE: otsuthresh [-g graph] infile outfile
USAGE: otsuthresh [-help]

-g ..... graph .......... graph specifies whether to generate a
......................... histogram graph image displaying the
......................... location and value of the threshold;
......................... choices are: view or save;
......................... default is no graph

PURPOSE: To automatically thresholds an image to binary (b/w) format using Otsu's technique.

DESCRIPTION: OTSUTHRESH automatically thresholds an image to binary (b/w) format. It assume the histogram is bimodal, i.e. is the composite of two bell-shaped distributions representing the foreground and background classes. The Otsu appoach computes the Between Class Variance from the foreground (above threshold data) and background (at and below threshold value) for every possible threshold value. The optimal threshold is the one that maximizes the Between Class Variance. This is equivalent to finding the threshold that minimizes the overlap between the two bell-shaped class curves.

ARGUMENTS:

-g graph ... GRAPH specifies whether to generate a graph (image) of the histogram, displaying the location and value of the threshold. The choices are: view, save and none. If graph=view is selected, the graph will be created and displayed automatically, but not saved. If graph=save is selected, then the graph will be created and saved to a file using the infile name, with "_histog_otsu.gif" appended, but the graph will not be displayed automatically. If -g option is not specified, then no graph will be created.

NOTE: It is highly recommended that the output not be specified as a JPG image as that will cause compression and potentially a non-binary (i.e. a graylevel) result. GIF is the recommended output format.

REFERENCES: see the following:
http://www.ph.tn.tudelft.nl/Courses/FIP/noframes/fip-Segmenta.html
http://homepages.inf.ed.ac.uk/rbf/CVonline/LOCAL_COPIES/MORSE/threshold.pdf
http://www.cse.unr.edu/~bebis/CS791E/Notes/Thresholding.pdf
http://www.ifi.uio.no/in384/info/threshold.ps
http://www.supelec.fr/ecole/radio/JPPS01.pdf

Mathematics Background (256KB PDF)

CAVEAT: No guarantee that this script will work on all platforms, nor that trapping of inconsistent parameters is complete and foolproof. Use At Your Own Risk.


EXAMPLES


Otsu Thresholding Of Various Images

Pictures Were Obtained from:
blood.jpg from http://www.istanbul.edu.tr/eng/ee/jeee/main/pages/issues/is62/62008.pdf
fingerprint.jpg from http://www.istanbul.edu.tr/eng/ee/jeee/main/pages/issues/is62/62008.pdf
flower.jpg from http://www.jhlabs.com/ip/blurring.html
house.jpg from http://en.wikipedia.org/wiki/Otsu's_method
kanji.jpg from http://www.measurement.sk/2004/S1/Yong.pdf
parts.gif from http://www.ph.tn.tudelft.nl/Courses/FIP/noframes/fip-Segmenta.html
rice.jpg from http://www.istanbul.edu.tr/eng/ee/jeee/main/pages/issues/is62/62008.pdf
tank.jpg from http://stinet.dtic.mil/cgi-bin/GetTRDoc?AD=ADA464347&Location=U2&doc=GetTRDoc.pdf
textsample.jpg from http://signal.ece.utexas.edu/seminars/dsp_seminars/01fall/211_seeger_mf.pdf
lena2g_edge1.jpg was created using the IM function -edge 1

Original Image

Thresholded Image

Histogram



See A Comparison Of Each Image Against Each Thresholding Technique



What the script does is as follows:

  • Converts the image to grayscale
  • Generates the histogram
  • Computes the normalized histogram counts by dividing each bin count
    by the total pixels in the image
  • Generates cumulative arrays for each bin from the normalized count
    and from the product of normalized counts times bin graylevel
    starting from each end of the histogram. These are called histogram
    moments zero and one
  • Computes the cumulative histogram mean arrays from these moment arrays
  • For each possible bin, finds the between class variance from the
    two cumulative mean arrays and chooses the threshold with the
    largest between class variance
  • Thresholds the image at the graylevel corresponding to the selected bin