Having learnt about cropping, we would now want to continue our discourse on face alignment. The cropping is to happen on the rotated images that we generated in one of our previous sections. So, ideally the implementation for image rotation and cropping are part of the same code snippet/function/module. However, for the purpose of illustrating and focusing on scaling/cropping separately in this section, we will first demonstrate the working of the same independent to the image rotation code. Then, in the next section, we integrate them together so that you get an idea of how these two operations can be coupled together to complete our facial alignment pipeline.
The main problem that we are going to deal with in this section is how to come up with suitable parameters that help us define our ROI. Note that in contrast to the previous section, we have a definite goal for the ROI in mind. We are cropping images so that the unnecessary background distractions...