Discussion of Algorithm
From the pictures we can clearly see that humans do focus more on the high spatial frequencies close up and low spatial frequencies far away. This fact is particularly evident in the dog/owner, heads/tails penny and the cat dog. While the images are certainly interesting to look at, there is also clearly some corruption. With the exception of the cat dog, it is easy to pick up some peculiarity in one or both of the images.The algorithm is clearly an interesting one, however, the fact remains that it is only truely effective when the two pictures being blended are incredibly similar or happen to compliment each other perfectly. The adjustment of the width of the Gaussian and the cutoff depth can be adjusted per image pair to achieve the best results. Particularly, sigma from the Gaussian can be increased to add more focus to the high frequency image and remove focus from the low frequency image. In other words, the frequency cutoff is a lower frequency when sigma is increased, so the resulting image has more from the high frequency image and less from the low frequency one. Even with carefully chosen pictures it is difficult to get one that can completely fool the human eye, especially with the up close image. Either the high or low frequency image can be grayed out to additionally adjust the balance. For example, the Cat Man picture looks much more like superimposed images as opposed to a hybrid images.
Even in the original paper the images are not perfect and the studies where humans were only able to deduce one image occured when the images were shown only for a short time. The results are certainly cool and entertaining, but to truely create hybrid images, the algorithm must be modified to exploit more quirks of the human visual system.