registration of images with different focuses

See more:

C#

image-processing

I've tried to write a code in order to do image registration of several images (from the same scene) that the only difference between them is the location of the focus at image. In the code I wrote there are the usual stages for an image registration: Extracting and matching similar features between the two input images(using SURF for extracting and correlation to match), and RANSAC to remove matching errors.

However, the result I get has errors in it - In cases of registering images with small difference between the images' focuses location there can be seen some kind of blur and shrinking of the objects. In cases with big difference between the images' focuses location I get a total mess, as if the program is trying to pin force objects in the images on each other.

Anyone has an idea how to improve it? I'm writing the code in C# using AForge, Accord and I can also use Emgu-CV, so you can suggest me to use functions from these libraries if you find it necessary / useful.

Thanks in advance.