Intensity/Entropy Scatterplot

Reading Chapter 4 of Machine Vision.

Below is a result from my implementation of the scatter plot described in section 4.3.2. My efforts in writing an entropy filter came out of the description of this graph. Davies does not mention an entropy filter explicitly and doesn't describe how one might calculate "intensity variation". I'm not certain that my graph is generating the same thing that he describes.

The X axis the the intensity value and the Y axis is the entropy value. The intensity and entropy are found for each pixel and used to increment a cell in the scatterplot.

Davies talks about using this to determine thresholding values, but I haven't studied what he wrote very closely yet. I got too excited thinking about how I might implement the entropy filter.

Things I want to do with this:

• Allow the user to select a point and have the software automatically find a Gaussian that describes the data around that point.
• Use a SVM to find linear separations of the clusters.
• Map the groupings defined by the SVM or Gaussian mixtures back to the image with each category getting a unique color. I'm curious if the groupings correspond to image features or distinctive areas.

See previous posts for the original image.