A Brief Introduction to ICM

Images get corrupted by many things, from noisy transmissions (e.g., from Mars rover to NASA) to dirty scanner glass and camera lenses. In such situations, we want to restore the image to its original or proper state as best we can. One method for doing this is called “Iterated Conditional Modes.” We won’t go in to all of the rich theory behind it, but the name does tell us two things. It’s iterated, so it will probably involve a loop. And by “mode” we mean finding a maximum (or actually a minimum, in our case).1 In short, we’ll be iteratively optimizing something within a loop, and that something is our best guess of what a corrupted image pixel should be. In this lab, we are going to be concerned with binary images. That is, images whose pixels are either on or off, black or white. Thus, the only noise that is possible is whether a black pixel (or bit) was flipped to a white one, and vice versa. When these images are huge (very high resolution telescope scans of the sky, or Google-scale satellite images of the earth), distributed parallel processing can make the job much faster.