The basic concept of representing an image by a relatively small number of features can be used for more than just classification. For example, we can also use it to find similar images to a given query image (as we did before with text documents).
We will compute the same features as before, with one important difference: we will ignore the bordering area of the picture. The reason is that, due to the amateur nature of the compositions, the edges of the picture often contain irrelevant elements. When the features are computed over the whole image, these elements are taken into account. By simply ignoring them, we get slightly better features. In the supervised example, it is not as important, as the learning algorithm will then learn which features are more informative and weigh them accordingly. When working in an unsupervised fashion, we...
