think about storing 1000 images. They can be just summed together like 1000 2d functions (function value represents color). Now suppose you see person 53. Your brain just forms the product of the image you see with the sum. Image 53 could stand out like in the case of Fourier series, since other images fall out due to (approximate) orthogonality (actually the difference level comparing to a threshold). This kind of retrieval (correlation based) is easy to implement. On the other hand if we add 1000 2d binary images we will need much more capacity for storage and/or multiplication, with the gain of unnecessary precision. Of course this makes HDTV girls look more attractive to you guys than the real persons. In the case of Obama he needs to store more than 100,000 faces....