I am a novice user, so please bear with me! I have a set of random points which are color-coded by some function. The point distribution is fairly dense so I should be able to see a smooth color distribution, but I see "choppy" discrete points. I use the following for processing:

g0 = Graphics[{PointSize @ .003, Opacity[0.075], Point[pts, VertexColors -> vc]}];

where pts is an array of 2D points (with (x,y) coordinates) and vc is some functional distribution of colors. The choice of PointSize essentially allows me to cover up the spaces that separate the points -- when PointSize is too small, one sees dots on a blank background, but when PointSize is too large, one sees an ugly, choppy color distribution, although the space separating the points is now covered.

I use Opacity to smooth out the choppiness, and for the most part it works. However, I would like to have a more elegant/exact control of this process. I would like to apply a radially symmetric filter function of sorts to these data points such that the color value at the center of each element of pts is vc and drops to some value (say, zero) at the edge of the radius of size PointSize. For the sake of argument let's just say the filter function at each point is vc * Max[0 , 1 - Abs[r/0.003 ]], where 0 <= r <= 0.003. This filter function should interpolate the colors in between points smoothly and provide a nicer looking field (it will look as if the points were first projected onto a uniform grid and then displayed smoothly).

Note that processing speed is somewhat important because I'm visualizing unsteady field of hundreds of thousands of points per timestep

Hopefully I'm clear enough with my request.

I'm including a side-by-side image: Left uses PointSize @ .005 (exaggerated to make my point); Right uses PointSize @ .005 and Opacity[0.07]. The problem with Opacity is loss of sharpness.

enter image description here

The "data" is at https://drive.google.com/file/d/0BzCeMgfPKi0vd0dCT2lRWTEzd0E/view?usp=sharing

The code snippet to extract the correct "data" and plot it with a color scheme that's close enough to my own is given below (setting Opacity to 1 shows the choppy version):

cf = ColorData["LightTemperatureMap"];
dat = Transpose[Import["data", "Table", "HeaderLines" -> 1]];
pts = Transpose[{dat[[1]], dat[[2]]}];
vc = cf /@ Rescale[Clip[-dat[[4]], {-50, 50}]];
g0 = Graphics[{[email protected], Opacity[0.1], Point[pts, VertexColors -> vc]},
         Axes -> False, PlotRange -> {{-.35, 2.15}, {-.8, .8}}, ImageSize -> {1200}];
  • 1
    $\begingroup$ Sample data would be greatly appreciated. $\endgroup$
    – C. E.
    Commented Mar 30, 2016 at 6:13
  • $\begingroup$ @ C. E. I'm adding the link to the data, as well as the code snippet, to my original question $\endgroup$
    – ASR
    Commented Mar 30, 2016 at 8:14

1 Answer 1


Perhaps an acceptable solution is to convert your plot to an image and apply very light filtering on it. At least to my eye, a median filter seems fit the bill here: it retains as much sharpness as possible, while "blending points" to form more uniform swaths

Using your data (notice that I removed your PlotRange option from Graphics for convenience in posting the pictures here, but it has no effect on the image generation):

g0 = Graphics[
   {[email protected], Point[pts, VertexColors -> vc]},
   Axes -> False, PlotRange -> All, ImageSize -> 1200
img = Image@%;
GraphicsRow[{img, MedianFilter[img, 4]}, ImageSize -> Full]

Below are the results. Your original Graphics object is on the left; the filtered image on the right (click on the image to open a larger version for better viewing).

two figures linked

Here is a zoomed-in detail of the filtered image. If you need higher-resolution results, you can increase ImageSize within Graphics before rasterization by Image.

zoomed in

  • $\begingroup$ Unfortunately, this approach doesn't work well. I've replaced the data file with a new one, representing the flow at a later time. The post-processed filtering approach is still spotty, or too smeared out if one uses a larger value for MedianFilter. The best approach would be to apply the "filter" to the points themselves. The "filter" acts like a Radial Basis Function (RBF) for interpolation, so the color values from neighboring points superimpose linearly to give a new color value (hence leading to a smooth color distribution). Opacity[] essentially behaves as a piecewise-constant RBF. $\endgroup$
    – ASR
    Commented Mar 30, 2016 at 23:03

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.