Notice that in Figure 18, the frequency response of each window has one main lobe at the center, and then multiple sidelobes at a lower response. To distinguish two signals that are close in frequency, we want the main lobe to be as narrow as possible. If our two frequencies are so close that they are both inside the main lobe, we cannot distinguish them. They will both show up in the same FFT bins. In this case, we don't care as much about the side lobes.
For the examples shown in Figure 18, the one with the narrowest main lobe (most selective) is the rectangular window. The Kaiser window comes in second. Note that the Kaiser window has an adjustable parameter beta, which you can use to tune the window to be more or less selective.
