Do you create these images from some form of parametric math, or do you have them "as is" as plain bitmaps? In the first case, you could just use so called "supersampling", which basically means: Render your image larger than it should be in the end, and downscale it with a nice filter. GR32 brings some very nice scaling filters along. In the latter case, you have a much much more serious problem, which cannot be solved this easily - especially not with affecting the image as a whole in a manner that it becomes blurry in all places.
A first step into the direction of an "intelligent" blur-filter is to use a crude form of edge-detection. You could, for example, filter the whole image using a
Sobel operator, and use the outcome as a mask that defines where to blur by how much. The blurring could be a gaussian filter with it's radius controlled per pixel by this mask.
"When one person suffers from a delusion, it is called insanity. When a million people suffer from a delusion, it is called religion." (Richard Dawkins)