Color Comparison

Author: Bill O'Connell (woc at attbi.com)
History: 2002/03/08: First version
2002/07/31: Added Regression option
2002/09/23: New features
2004/03/09: Updated to work with byte masks used in ImageJ 1.32c and later
Requires: ImageJ 1.32c or later
Source: ColorComparison_.java
Installation: Download ColorComparison_.java to the plugins folder and compile it using Plugins/Compile and Run. There will be a new "ColorComparison" command in the Plugins menu the next time you restart ImageJ.
Description:
Color comparison of two 8-bit identically dimensioned gray scale images.
Each pixel has an intensity in image A and in image B. Plot each point
on a plain with image A intensity as abcissa and B intensity as ordinate.
ColorComparison allows the user to compare Image1 with Image2 intensities
using one of the methods below -

  1) Polar Comparison: In polar coordinates, a pixel's location is given by magnitude and
  angle. The color image shows pixel magnitude (or image A or image B intensity) as
  intensity. Polar angle is shown as hue. With positive valued images, pixel angle is 
  between 0 and 90 degrees. Pixel angles are mapped to any continuous swatch of a color
  wheel where
      R=0, G=120, B=240 degrees.
   
  2) Percentage: p = 100 * (B intensity) / (A intensity + B intensity)
  may be used to specify hue. With methods 1) and 2) the mapping is isomorphic -
  all information in the two gray scale images is retained in the color image. 
  ColorComparison also generates a histogram of pixel pair angles(0 to
  90 degrees) or percentages(0 to 100%) sorted into a user specified number of bins.

  3) Regression: A linear regression and standard error of regression are calculated
  predicting Image2 from Image1 pixel intensities. The hue of each pixel in the color
  image expresses the residual, the difference between the predicted and actual
  Image2 intensity divided by the standard error of regression.
  
The range of differences mapped by the color image is set by the number of histogram
bins specified. For instance, sepecifying 10 bins sets a range of -5 to +5 standard
errors.

The hue of pixels outside this range are clamped to the max or min standard error.
After the color image is generated, the user is allowed to draw ROIs on Image1.
The mean residual within the ROI and its probability are displayed in a table.
If the histogram uses weighted intensities the mean is calculated using weighted
intensities.

if HighlightPixels is checked, the program highlights all color image pixels with
values outside regression confidence limits at the specified probability level alpha.
Percentages of pixels above and below the confidences limits are displayed in the
Results window. Pixels below the specified Theshold Intensity are excluded, and
percentages are weighted by Image1 intensities if the weight histogram box is checked. 

|Plugins | Home |