NEXT | BACK | INDEX

PREDICTOR ALGORITHMS

Predictor algorithms try to "predict" the value of a pixel, and only store the difference between the predicted value and the actual value. There are many variations on this basic theme.

The simplest algorithms use already known nearby pixels to predict the next pixel value. If the image is rastered from left to right, and then from top to bottom imgCIF / CBF predictors defined 4 predictors for an unknown pixel X, based on up to 4 known pixels A, B, C, D. 

              C  B  D

              A  X
The value stored after the predictor is:
  1. X' = X - A

  2. X' = X - Int((A + B) / 2)

  3. X' = X - Int((A + B + C) / 3)

  4. X' = X - Int((A + B + C + D) / 4)

The aim is for the probability of values in X' to be much more peaked than for X.

For most X-ray, neutron, and electron diffraction data, predictors based on nearby surrounding values will be very close to optimum owing to the high counting statistics noise, and detector point spread functions leading to correlated noise.