The compositing equation describes how foreground pixels combine with the background even when taking a photo of a scene. In general, you should assume that the foreground can be partially transparent as controlled by its alpha (but not refractive and not exhibiting colored transparency).
In fact, when a camera images even an opaque object, camera pixels will generally see a mixture of foreground and background at the boundary of the foreground silhouette. This occurs because the camera pixels have area, and the foreground silhouette boundary generally does not line up to cover pixels completely, so a partial mixture of foreground and background is observed at these pixels. This effect is actually exaggerated by any defocus in a real imaging system.
The given filter B, though shown as a 3x3 filter, is essentially a 1x3 filter padded with zeros. The second part of the question asks you to compare this to using a "full" 3x3 filter. For reference, examples of "full" filters are the mean and Gaussian filters discussed in lecture.