As part of the previous chapter's work, we wrote copyRect() as a copy operation that limits itself to given rectangles of the source and destination images. Now, we want to apply further limits to this copy operation. We want to use a given mask that has the same dimensions as the source rectangle. We shall copy only those pixels in the source rectangle where the mask's value is not zero. Other pixels shall retain their old values from the destination image. This logic, with an array of conditions and two arrays of possible output values, can be expressed concisely with the numpy.where() function that we have recently learned.

Let's open rects.py and edit copyRect() to add a new argument, mask. This argument may be None, in which case we fall back to our old implementation of the copy operation. Otherwise, we next ensure that mask and the images have the same number of channels. We assume that mask has one channel but the images may have three channels (BGR). We...