SoFirstOrderStatisticsProcessing Class Reference
[Texture Filters]

SoFirstOrderStatisticsProcessing image filter More...

#include <ImageViz/Engines/ImageFiltering/TextureFilters/SoFirstOrderStatisticsProcessing.h>

Inheritance diagram for SoFirstOrderStatisticsProcessing:

Public Types
enum	MeasureType { MEAN = 0, VARIANCE = 1, SKEWNESS = 2, KURTOSIS = 3, CONTRAST = 4, VARIATION = 5, ENERGY = 6, ENTROPY = 7 }
Public Member Functions
	SoFirstOrderStatisticsProcessing ()
Public Attributes
SoSFEnum	measureType
SoSFImageDataAdapter	inImage
SoSFInt32	kernelSize
SoImageVizEngineOutput < SoSFImageDataAdapter, SoImageDataAdapter * >	outImage

Detailed Description

SoFirstOrderStatisticsProcessing image filter

SoFirstOrderStatisticsProcessing image filter computes first order statistics.

The SoFirstOrderStatisticsProcessing filter is dealing with first order statistics. It creates a result image where values on pixels are only function of values of this pixel in the initial image and its neighbourhood of a wanted size. In order to calculate this values, we calculate local histograms (array of number of pixels per value in the neighbourhood). In the following equations, $p(n)$ corresponds to the probability to have a pixel with value $n$ (it's the number of those pixels divided by the total number of pixels in the neighbourhood).

Reduced image

For ENTROPY and ENERGY types, we must reduce the number of grey levels in order to calculate more significant statistics. We have two parameters to reduce images : having 4 or 8 classes, and with an equal repartition or not. For an equal repartition, we use the quantiles :

$q_t=inf\Big\{q,\sum_{r\le q}p(r)\ge t\Big\}$

So, for 4 classes, we use :

$q_\frac{1}{4},q_\frac{1}{2},q_\frac{3}{4}$

and for 8 classes :

$q_\frac{1}{8},q_\frac{1}{4},q_\frac{3}{8},q_\frac{1}{2},q_\frac{5}{8},q_\frac{3}{4},q_\frac{7}{8}$

In the other case (no equi-repartition required), we use mean and standard deviation. So the delimiters for 4 classes are :

$(m-\sigma),(m),(m+\sigma)$

For 8 classes, we calculate the mean $m$ , then we obtain two subsets where we calculate means $m1$ and $m2$ , and standard deviations, we have in that case :

$(m_1-\sigma_1),(m_1),(m_1+\sigma_1),(m),(m_2-\sigma_2),(m_2),(m_2+\sigma_2)$

This values delimitates the classes and the value given to the pixels which belong to one class is the middle of the segment.

FILE FORMAT/DEFAULT

measureType	MEAN
inImage	NULL
kernelSize	3

Library references: contrast energy entropy kurtosis mean skewness variance variation

Deprecated:: Deprecated since Open Inventor 9800
Replaced by SoLocalStatisticsProcessing

Member Enumeration Documentation

enum SoFirstOrderStatisticsProcessing::MeasureType

Enumerator:

MEAN	The MEAN type gives to pixels the mean value in its neighbourhood: $mean=m=\sum_{n}n\cdot p(n)$ .
VARIANCE	The VARIANCE type gives to pixels the variance value in its neighbourhood: $variance=\sigma^2=\sum_{n}(n-m)^2\cdot p(n)$ .
SKEWNESS	The SKEWNESS type gives to pixels the skewness value in its neighbourhood: $skewness=\frac{\sum\limits_{n}(n-m)^3\cdot p(n)}{\sigma^3}$ .
KURTOSIS	The KURTOSIS type gives to pixels the kurtosis value in its neighbourhood: $kurtosis=\frac{\sum\limits_{n}(n-m)^4\cdot p(n)}{\sigma^4}-3$ .
CONTRAST	The CONTRAST type gives to pixels the contrast value in its neighbourhood: $contrast=\frac{max(n)-min(n)}{max(n)+min(n)}$ .
VARIATION	The VARIATION type gives to pixels the variation value in its neighbourhood: $variation=\frac{m}{\sigma}$ .
ENERGY	The ENERGY type gives to pixels the energy value in its neighbourhood in the reduced image: $energy=\sum_{n}p(n)^2$ .
ENTROPY	The ENTROPY type gives to pixels the entropy of its neighbourhood in the reduced image: $entropy=-\sum_{n}p(n)\cdot\log p(n)$ . See also: SoCrossCorrelationProcessing2d