Followup: Density Mapping

[Blammo]

I would be interested  to hear about your speed tests, "how many points 
take so long to generate" sort of stuff.

Your overlays look pretty good.  With an integration into MapServer, 
there would also bthe possiblity of adding in semi-opaqueness to the 
overlay output.  This was one major area that gave us fits with the 
overlays that we tried.

Thanks

bobb


Bill Binko wrote:

>Hi everyone.
>
>A few weeks ago, I asked a question about density mapping, and got
>several responses, including several very helpful pointers.
>
>I thought I would write back with where the progress I've made, along
>with another request for feedback.
>
>Here are three images that show some of the density maps I've been able
>to make on the fly:
>
>http://m.mapshine.com/images/dazzlehX6EfG.png
>http://m.mapshine.com/images/dazzlesSqxTl.png
>http://m.mapshine.com/images/dazzlePjFMQF.png
>
>I'm sure there are statistics professors out there who will roll their
>eyes at my approach, (and please, if I'm making any horrific mistakes,
>please let me know) but the proof-of-concept I developed is performing
>quite well in terms of speed.
>
>The proof-of-concept works like this (in PHP/Mapscript):
>
>1) Read a Standard Deviation in geographic units (1/3 mile for the
>images above)
>2) Generate a "stamp" image that is a gradient that starts at 100%
>opaque at the center and "fades out" to 0% opaque at the center using
>the Parzen Window Gaussian Kernel
>equation:http://en.wikipedia.org/wiki/Parzen_window where "x" is
>actually the distance from the center of the stamp
>3) Create a new image to lay over the Mapserver generate map with the
>same size, and "ImageAlphaBlending()" set to true.
>4) Query the map layer using "whichShapes()" to get the points on the
>layer we're interested in
>5) Iterate through all the points on the layer and draw the "stamp"
>image into the new image.  Because of GD's Alpha Blending, these
>"stamps" build up effectively performing the Summation in the Parzen
>Window document above.
>6) Walk through the entire image (one point at a time) and map the 0-127
>alpha value of each point onto table of images (just to make it look
>good :)
>
>I have thought about how this type of functionality could make it back
>into the core Mapserver, and I actually think it could be fairly simple.
>However, this is obviously a proof of concept, and I'm not pushing for
>this... I just wanted to update everyone.
>
>To take this same approach with the C Mapserver code, we'd need to do
>the following:
>
>1) Implement semi-transparent symbols per danmo's suggestion (right now,
>we only have alpha transparency at the layer level, not the symbol
>level)
>2) Setup a hook so that user code can create the png "stamp" after the
>layer's requested extent is known and before any drawing is done
>3) Setup a hook where the user can get access to the raw layer image for
>post-processing.
>
>I am not 100% sure that the approach I've taken is completely valid.
>However, I am quite sure that is far faster than the alternatives I
>faced which were a) spawning out to GRASS/R to generate a large raster
>or b) doing a distance calculation + a gaussian equation for each pair
>of pixel/Point in the extent!  
>
>I look forward to your responses.
>
>Bill
>
>  
>