What is a Hybrid Image?
A hybrid image is the combination of two other images, such that the resulting one appears to be one thing when viewed up close and appears to be another, when viewed from a distance. To begin, we need two images that are roughly aligned. We then filter both images to create a Gaussian and Laplacian Pyramid for each each image. We then use these pyramids to reconstruct a single image, such that the resulting image has the low frequencies of one image and the high frequencies of the other. Try viewing the image in the upper left hand corner. When close to the screen it should appear to be a dog with a rather large shadow or halo of some sort. From a distance, however, the dog will dissapear and a gorilla should take its place.
How does it Work?
Visual researchers suggest that hemans have multi-bandwith means for interpreting images. At a fast glance or from a distance, humans only make out the general large scale structure of an image. To gather this "global" understanding, humans only utilize the low spatial frequencies they recieve from an image or real world scene. Upon a closer look, we search for fine details (high spatial frequencies). Taking advantage of this phenomena of human vision, by removing high spatial frequencies from one image and replacing them with those from another, we make a new image which depending on how carefully and from what distance it is examined, it can appear to be one of two things. Other factors can also effect which spatial frequencies we analyze, such as our task in what we are trying to anaylze. For more information, follow the link to Oliva, Torralba, and Schyn's paper.