Resampling Methods

There are very many ways to resample a digital picture—to redraw it using more or fewer pixels, either permanently or just on-screen. These methods differ in how they decide on values for the pixels in the new version of the picture, and in how many pixels in the original are analyzed for such decisions. There is no "silver bullet" – each method has its advantages and disadvantages. It all depends on what is being resampled, and why.

One way in which resampling methods differ is the "sharpness" of the output picture. When you are shrinking pictures, some methods, like bicubic and supersampling, lead to a mildly blurry image. Thus it is a good idea to mildly sharpen after resampling in such cases.

  • Nearest neighbor – the simplest and fastest method; it does not use any interpolation, but rather evaluates each of the original's pixels on their own; a bad choice for photos, but irreplaceable for technical drawings with hairlines.
  • Bilinear – the simplest kind of interpolation; uses the relative sum of the four nearest pixels; fast and generally good when shrinking a picture.
  • Bicubic – relatively advanced interpolation; uses the 16 nearest pixels; interpolates values along a cubic curve; suitable for both stretching and shrinking (if the picture is sharpened afterwards).
  • Hermite – another type of interpolated curve; uses the four nearest neighboring pixels.
  • Bell – gives a very "soft" image; useful for pictures with noise.
  • Mitchell – an excellent combination of speed and quality; uses the 16 nearest pixels; has a "self-sharpening" effect.
  • Lanczos – processor-intensive; pixels are interpolated using a special curve simulating the real dissemination of information; 36 pixels from the original are used per output pixel; has a strong "self-sharpening effect"; most useful when stretching pictures; can cause unaesthetic grid-like artifacts during shrinking due to the sharpening effect.
  • Supersampling – designed only for shrinking pictures; uses the weighted average of all the pixels that are lost during shrinking. Generally gives the best results for photos, because it works with all pixels in a photo. Can suffer from unsharpness, but this can be solved by mild sharpening afterwards.