CSE

Description

Edits (modifies) class signatures prior to maximum-likelihood or parallelepiped classification. The analyst can modify the encoding value, the class threshold and/or the class 'a priori' probability, as well as any layer mean, standard deviation, and lower and/or upper parallelepiped limits.

Parameters

Name	Type	Length	Value range
InputSIG: Input class signature segment *	SIG port	1 - 1
OutputSIG: Output class signature segment *	SIG port	1 - 1
Gray-Level Value	Integer	0 - 1	1 - 254
Gaussian Threshold	Float	0 - 1
Class Bias	Float	0 - 1
Layer to Modify	Integer	0 - 1
Mean Gray Level	Float	0 - 1
Standard Deviation	Float	0 - 1
Lower Parallelepiped Class LImit	Float	0 - 1
Upper Parallelepiped Class Limit	Float	0 - 1

* Required parameter

Parameter descriptions

InputSIG: Input class signature segment

Specifies the class signature segment (type 121) to modify.

OutputSIG: Output class signature segment

Specifies the output class signature segment (type 121) to receive the modified signature.

Gray-Level Value

Specifies an integer between 1 and 254 that represents the class in the output classification theme map created by the Maximum Likelihood Classifier (MLC).

If no value is specified, no change will occur.

When maximum likelihood classification is used, the integer value 0 is reserved for the null class and the integer value 255 is reserved for pixels that fall into multiple classes. Therefore, the values 0 and 255 are invalid

Gaussian Threshold

Specifies the Gaussian threshold of the signature segment. The threshold is the radius, expressed in standard deviation units, of a hyperellipse surrounding the mean of the class in the feature space. If the threshold is increased, the size of the class hyperellipsoid also increases. This results in the capture of more feature space at the expense of the null class. The capture is performed at classification time for the class that you are editing.

If no value is specified, no change will occur.

This parameter's value is a real number expressed in standard deviation units. It is used only for the maximum likelihood classification.

Class Bias

Specifies the relative weighting among classes. If all the classes have equal biases (relative "a priori" probabilities), these relative weights must remain equal to each other. The "a priori" probability biases the maximum likelihood class assignment for points that overlap in the feature space. Large relative biases capture more of the feature space at the expense of overlapping classes with lower biases. This is done at classification time for the class that you are editing.

If no value is specified, no change will occur.

This parameter is used only for the maximum likelihood classification.

Layer to Modify

Specifies the input channel for which you wish to modify the mean, standard deviation, lower limit, and/or upper limit.

This parameter must be specified if the mean, deviation, lower and/or upper limit are to be modified. If no value is specified, no changes will occur.

Mean Gray Level

Specifies the mean of the class signature segment for the specified image channel.

The MEAN is a real number. The Layer to Modify (DBIC) must also be specified.

If no value is specified, no change will occur.

Standard Deviation

Specifies the standard deviation of the class signature for the specified image channel.

The standard deviation is a real number. The Layer to Modify (DBIC) must also be specified.

If no value is specified, no change will occur.

Lower Parallelepiped Class LImit

Specifies the lower class limit of the class signature for the specified image channel.

The lower class limit specifies, in standard deviation units, the distance from the class mean to the lower boundary of the parallelepiped for a specified image channel (DBIC). The lower class limit is used only for the parallelepiped classification.

The lower class limit is a real number. The Layer to Modify (DBIC) must also be specified.

If no value is specified, no change will occur.

Upper Parallelepiped Class Limit

Specifies the upper class limit of the class signature for the specified image channel.

The upper class limit specifies, in standard deviation units, the distance from the class mean to the upper boundary of the parallelepiped for a specified image channel (DBIC). The upper class limit is used only for the parallelepiped classification.

The upper class limit is a real number. The Layer to Modify (DBIC) must also be specified.

If no value is specified, no change will occur.

Details

CSE allows you to modify the output encoding value, class threshold, relative "a priori" probability (BIAS) and for a channel, the mean, standard deviation, and the lower and upper parallelepiped limits.

Modifying these values is especially useful if you have more reliable values for the mean and standard deviations for a class, obtained from published results.

The VALU (Gray Level Value) parameter is the encoding value used for the output classification theme map created for the maximum likelihood classification. The values that you select and modify depend on the type of classification for which the signature is to be used. A signature segment used for the maximum likelihood classification describes a hyperellipsoid in feature space for a class. Each hyperellipsoid has a particular shape (long, thin, or round), size, and position. Increasing the threshold (THRS) increases the hyperellipsoid size, allowing more pixels to fall in that class. Changing the mean (MEAN), shifts the center of the hyperellipsoid along a specified image plane in feature space (DBIC) without changing its shape or size. Modifying the standard deviation for a channel changes the shape of the hyperellipsoid in the specified channel feature space plane.

The BIAS provides relative weighting between the maximum likelihood classes. If all the classes have equal biases (relative "a priori" probabilities), these relative weights must remain equal to each other. This "a priori" probability biases the maximum likelihood class assignment for points that overlap in the feature space. Large relative biases capture more of the feature space at the expense of overlapping classes with lower biases. This is done at classification time for the class that you are editing.

A signature segment used for the parallelepiped classification describes a multi-dimensional box or parallelepiped in feature space. The shape of a parallelepiped, for a particular image plan (DBIC), can be changed by altering the standard deviation or the parameters LOLIM (Lower Class Limit) or UPLIM (Upper Class Limit), or both. This is illustrated in the following equations:

low gray-level = mean(i) - (stdv(i)*lolim(i)+0.5)
hi  gray-level = mean(i) + (stdv(i)*uplim(i)+0.5)

As with the hyperellipsoid, changing the mean (MEAN) shifts the center of the parallelepiped along a specified image plane in the feature space (DBIC) without changing its shape or size.

You can change any or all of the signature values at once. You must, however, set the appropriate parameters only for those signature values that you want to modify; the settings for the parameters of those signature values that are not being changed must be left blank. For example, to change the threshold, set the appropriate THRS value and clear the settings indicated for VALU, BIAS, DBIC, MEAN, STDV, LOLIM, and UPLIM.

Use CSR to obtain a report on the current mean values, standard deviations, and so on for all channels of a signature.