FLTRNOI

Description

FLTRNOIS filters noise in an image channel or layer, using a registered noise-free reference image that is sampled at the same pixel size as the noisy image. The reference image can be from a different date, frequency band, and sensor.

Parameters

Name	Type	Length	Value range
InputRef: Reference Image Layer: Input reference channel(s) or layer(s) *	Raster port	1 - 1
InputNS: Noisy Image Layer: Input noisy channel(s) or layer(s) *	Raster port	1 - 1
Output: Output raster channel(s) or layer(s) *	Raster port	1 - 1
Port Settings: Resample mode	Raster port	1 - 1024	Nearest | Bilinear | Cubic Default: Nearest
Maximum Model Gain	Float	0 - 1	0.0 - 256.0 Default: 3.0
Kernel Size	Integer	0 - 1	1 - 100 Default: 7
Degree of enhancement	Float	0 - 1	0 - 10.0 Default: 1.0
Report	String	0 - 192	See parameter description

* Required parameter

Parameter descriptions

InputRef: Reference Image Layer: Input reference channel(s) or layer(s)

Specifies the image channel to be used as the noise-free reference. This must be an 8-bit unsigned image channel.

InputNS: Noisy Image Layer: Input noisy channel(s) or layer(s)

Specifies the noisy image channel or layer to filter. This must be an 8-bit unsigned image channel.

Output: Output raster channel(s) or layer(s)

Specifies the output channel(s) or layer(s) to receive the noise-filtered results.

This must be an 8-bit unsigned image channel.

Port Settings: Resample mode

Specifies the type of image resampling desired.

Maximum Model Gain

Specifies the maximum model gain (MAXGAIN) to be used during enhancement. Lower values reduce the tendency of the algorithm to introduce noise into water regions. A typical starting value is 3 (default value). If noise results in water regions, reduce the maximum gain value.

Kernel Size

Specifies the size of the linear kernels on which the cross-correlation modelling is performed. Correlation is done on a window size of ( 2 * KSIZE + 1 ) pixels.

A kernel size of less than 3 will produce noisier images. Kernel sizes greater than 7 will blur edges.

Degree of enhancement

Specifies the degree of enhancement. This parameter provides as output a linear combination of the noisy input and the filtered version. Values lower than 1 will be noisier.

Report

Specifies where to direct the generated report.

Available options are:

• LOG: generates a report on the terminal (default)
• <modulename.RPT>: appends a report to the module's RPT file
• DISK: generates a report on file "IMPRPT.LST"

Details

FLTRNOI performs a cross-correlation between the input reference image (FILI_REF) and the input noisy image (FILI_NS), and creates a horizontal and vertical kernel defined by the Kernel Size (KSIZE) parameter. As this kernel scans across the window, the merit of both the horizontal and vertical kernels are evaluated, and the correlation coefficients (b0 and b1) of the kernel with the best merit are used. When the b0 and b1 matrices are obtained, the output image is reconstructed using the reference image b0 and b1 matrices in the following manner:

ORIGINAL_OUTPUT = B0 + B1 * FILI_REF

The amount of smoothing while preserving edge detail in the output image can be controlled with the maxium model gain parameter (MAXGAIN), which sets the upper limit of the B1 coefficient of the kernel with the best merit. For example, setting MAXGAIN = 0 results in FILI_NS as output, with no enhancement applied.

The ENH (degree of enhancement) parameter controls the enhancement factor for the noise reduction filter:

OUTPUT = ENH * ORIGINAL_OUTPUT + (1 - ENH) * FILI_NS

An enhancement factor of 1 produces only the enhanced image. An enhancement factor is 0 produces only the noisy image. As the enhancement factor is adjusted from 0 -> 1, the amount of enhancement increases as the percentage of information from the noisy image decreases.