Why do we think a vague image after supersampling is better than an aliasing image?