Other Concepts

Resampling

In order to get the resulting image \(R\) when applying a transformation \(T\) to an image \(I\), we need values for points \((x,y)\in\mathbb{N}^2\). If we assume the original image is continuous, we could get these values as \(R(x,y) = I(T(x,y))\), but this is generally not possible as \(I\) only has values at integer indices, and \(T(x,y)\) could be any pair of real numbers. To get values for non-integer positions, we need some way of interpolating the nearby values.

In practice, the main interpolation methods are nearest-neighbor, bilinear and bicubic interpolation.

Fig 1. nearest-neighbor

Fig 2. bilinear

Fig 3. bicubic