实践中的插值

Nearest neighbor and its staircase

Bilinear, bicubic and Lanczos

Upsampling adds pixels, not information

Choosing a kernel

Which statement correctly distinguishes nearest-neighbor and bilinear interpolation when magnifying an image by a factor of two?

A $32 \times 32$ image is upsampled to $256 \times 256$ with a high-quality Lanczos kernel. What has been added to the image's information content?

A one-dimensional image has sample values 2 and 6 at adjacent integer positions. Bilinear interpolation evaluates the value exactly halfway between them. Compute that midpoint value.