Class LocalContrastNormalisation
Performs local contrast normalisation on a matrix of values where the matrix is assumed to be derived from an image.
Inherited Members
Namespace: AudioAnalysisTools.DSP
Assembly: AudioAnalysisTools.dll
Syntax
public class LocalContrastNormalisationMethods
| Improve this Doc View SourceComputeLCN(Double[,], Int32)
WARNING!!: This method implements a convolution and like all convolutions is very slow (unless it can be fully parellised); Consider using another noise normalisation method such as in the class NoiseRemoval_Briggs.
This method does local contrast normalisation. Typically LCN normalises by division only and is motivated by what is known to happen in the visual cortext. Every matrix element or pixel value is divided by the (scaled) standard deviation of pixel values in a local field centred on the pixel. Scaling affects the severity of the normalisation. There are several ways of doing LCN. That is can divide by the local sum of squares. Or can calculate the local z-score which effectively normalises by both subtraction and division. This method is based on formula given by LeCun. Python code at the bottom of this class is the actual code used by LeCun which appears to do something different. Wish I knew Python!.
Declaration
public static double[, ] ComputeLCN(double[, ] inputM, int fieldSize)Parameters
Returns
| Type | Description |
|---|---|
| Double[,] |