Noise sampling is the first and most important step in removing noise from an image or in matching the noise of one image in another image. Normally, this process is entirely automatic. For fine control, you can switch to Manual mode and adjust the samples using the Sampling controls group in the Effect Controls panel.

A noise sample should be a solid block of uniform color that clearly displays the noise pattern present in the image. The objective is to extract samples of pure noise, without any image features that the algorithm could misconstrue as grain. For example, extract samples from a piece of sky, a background wall, or an area of fleshtone. All samples should be selected from the normal range of the film, DV, or video stock. Avoid underexposed or overexposed areas lacking in information, especially areas where pixel values have been clipped to pure black or white. Within this normal exposure range, it’s best to select samples with various RGB values and colors—for example, one sample from a bright area, one from a dark area, and one from an area with midtones.

The number of samples in automatic mode is high to ensure that the algorithm has enough good noise data, even if finding good samples in a particular image is difficult. In addition, automatic mode may override the number of samples you’ve set if the effect can’t find enough good samples. You can vary the size of the samples in either automatic or manual mode; however, increasing sample size doesn’t guarantee better results, especially if the resulting samples include more substantial variations in RGB values. Sample size should be reduced if a particular image doesn’t contain sufficiently large areas of constant color values. Conversely, increasing the sample size may give better results if the image contains large featureless areas.