This section will describe the kind of features that are extracted from the input images when we are trying to develop a face detection system. Since we have already stated that detecting faces in images is no trivial task for machines, you would expect the feature extraction step to be fairly complicated. Surprisingly, it is not so! As it turns out, extracting features is as simple as taking the sum of pixel values values over arbitrary rectangular sub-regions within images! One of the reasons for the popularity of the Viola-Jones detector is its simplicity. Let us learn a bit more about these features.
The feature-set that we are going to talk about in brief here are known as Haar-like features, or simply Haar features. They are nothing but the difference of the sum of intensity values within rectangular regions. If you look at the figure shown in a while, you will notice that each of the four given arrangements of rectangular regions (A, B, C, and D) have some clear and shaded...
