There are two problems that lie at the core of image understanding. The first is image classification , where we decide what class an image of a fixed size belongs to. It’s usual to work with a collection of images of objects. These objects will be largely centered in the image, and largely isolated. Each image will have an associated object name, using a taxonomy of classes provided in advance. You should think of catalog images of clothing or furniture. Another possible example is mugshot photos or pictures of people on websites (where the taxonomy is names). Judging by the amount of industry money pouring into image classification research, there are valuable applications for solutions.