Little Pittsburgh
This is a picture of downtown Pittsburgh from flickr that has been turned into a miniature. I think it looks quite fun. To get this result I implemented depth-of-field blurring based on an arbitrary polygon. The buildings of interest are outlined by a polygon. Then the areas further away from the polygon are then blurred progressively. Thus, the further away from the polygon, the blurrier. The program also increases the saturation of the image slightly to give it a better mini look.
Tweaking
The biggest tweaking of the algorithm is by how much the blur increases. In other words, how fast things far from the polygon of interest get blurred. Eventually I found a pretty good setting by using a 3x3 Gaussian blur with lambda 0.5. I then perform 20 depth steps who pixel length vary depending on the original width and height of the image. The key to working with an arbitrary polygon is using the "dilation" morph. However, the built-in dilation function (bwmorph(bw, 'dilate')) proved rather slow. Thus, I performed my own slightly modified dilation by shrinking the mask size, filtering, then resizing the mask back to its original size. Finally, how much saturation to add to the image had to be modified until I found a good value.
Little France
  
My own take on some often miniturized pictures of the Champs Elysees in Paris (left, right). Rather than take a horizontal depth-of-field, I chose a polygon that followed the road going into the distance. Thus, the center portions of the images are in focus while the sides and far back get blurrier. I personally like the one of the left; it looks like it could be a model.
Other Results