The two top figures show the response of an ideal low pass filter and (the center part of) its corresponding infinitely long impulse response. The bottom figures show the result of windowing the impulse response, i.e., cutting out the $N$ center samples and making the rest zero, which gives you a finite length impulse response that can be implemented. You can see that windowing obviously changes the frequency response, but the longer the window is made the better the approximation. However, note that the maximum approximation error (around the discontinuity) always has the same value, regardless of the window length. This is called the Gibbs phenomenon.