The first project, hybrid images, uses the high-frequency details of one image and low-frequency details of a second image to trick the human eye into seeing one thing up close and another far away. Spacial continuity matters (eyes should line up, main edges should line up, etc). We also use Gaussian/Laplacian pyramids to pick the optimal cutoff for the high and low pass filters (these cutoffs change per image). The Gaussian pyramid is created by blurring the image, then downsampling it (by a factor of 2) and repeating. The laplacian is created by taking the difference between each gaussian level (preserving only the detail/high-frequency information). In general, I found that the 4th gaussian level and 2nd laplacian level worked the best together; they provided the right level of blurriness for the first image, and a good level of detail for the second. This didn't work all the time, but it was a good baseline.

The results are as follows:
+ =
+ =
+ =
+ =
+ =

And, for the Doctor Who fans out there...
+ =