FTOUZI

Description

FTOUZI performs multiresolution adaptive filtering on image data. Primarily, you use FTOUZI on SAR data to remove high-frequency noise (speckle) while preserving high-frequency features, such as point targets, curvilinear features, and edges. The size of the adaptive filter is controlled by changes in the coefficient of variation and the gradient of the ratio edge detector for successive window sizes.

Parameters

Name	Type	Length	Value range
Input Layer: Input detected SAR image*	Raster port	1 - 1
Mask: Area mask	Bitmap port	0 - 1
Output: Output filtered image*	Raster port	1 - 1
Filter Size	Integer	0 - 1	11 | 13 | 15 | 17 | 19 | 21 | 23 | 25 | 27 | 29 | 31 | 33 Default: 11
Number of Looks	Float	0 - 1	1 - 100 Default: 1
Curve Detection Threshold	Float	0 - 1	0.0 - 10.0 Default: 0.5
Edge Detection Threshold	Float	0 - 1	0.0 - 10.0 Default: 0.5
Gradient Threshold	Float	0 - 1	0.0 - 10.0 Default: 0.1
Image Units	String	0 - 1	Amplitude | Power Default: Amplitude

* Required parameter

Parameter descriptions

Input Layer: Input detected SAR image

The image layers to filter.

Mask: Area mask

The bitmap that defines the area to process in the input raster. If no value is specified fot this parameter, the entire layer is used by default. With a bitmap mask, you must specify the bitmap segment that you want to use. All of the pixels within the specified segment, having a pixel value of 1, define the area to process.

Only the area under the mask is written to the output.

This parameter is optional.

Output: Output filtered image

The output channel or channels to which to write the filtered results. If the specified output file already exists, FTOUZI appends the filtered channels to the existing file. If the output file does not already exist, FTOUZI creates a new file.

Filter Size

The filter size of the square window, in pixels. The value must be an odd integer between 11 and 33. For example, to specify a window size of 11 x 11 pixels, type 11.

The default value is 11.

This parameter is optional.

Number of Looks

The effective number of looks of the image, which FSPEC uses to derive noise variance. The value of this parameter determines the amount of smoothing applied to the image.

The range of acceptable values is from 1 to 100. The default value of 1 is appropriate for single-look data and applies maximum smoothing. However, if in the file metadata NumLooks contains a value, that value will be used as the default.

This parameter is optional.

Curve Detection Threshold

The threshold of the curve-detection algorithm that determines whether a given pixel is part of a curved (or linear) feature.

Using higher values for this parameter preserves contours and curvilinear features, but reduces smoothing. Lower values increase the smoothing, but reduce preservation of contours or curvilinear features.

The default value of 0.5 is sufficient for most SAR images.

This parameter is optional.

Edge Detection Threshold

The threshold of the edge-detection algorithm that determines whether a given pixel is part of an edge.

Using higher values for this parameter preserves feature edges, but reduces smoothing. Lower values increase the smoothing, but reduce preservation of feature edges.

The default value of 0.5 is sufficient for most SAR images.

This parameter is optional.

Gradient Threshold

The maximum gradient of the coefficient of variation permitted between successive windows when determining the maximum isotropic area. Higher values increases the size of isotropic areas.

The default value of 0.1 is sufficient for most SAR images.

This parameter is optional.

Image Units

The format or type of radar image.

The image can be in Amplitude (AMP) or Power (POW). This information is usually in the image metadata. With a complex number a + bi, a corresponds to the real part (I) and b to the imaginary part (Q).

Details

The FTOUZI filter is a multicomponent, multiresolution, adaptive algorithm. Subareas within the processing window are identified as point targets, curvilinear features, homogeneous regions, edges, or isotropic regions. After a subarea type has been identified, iterative window expansion delineates the maximum extent of the identified feature. Each maximized area is filtered using the most appropriate window size.

Different filter sizes will greatly affect the quality of processed images. The Touzi filter removes SAR speckle, while preserving the spatial-signal variability (texture and fine structures). If the filter window is too small, the noise-filtering algorithm is not effective. If the filter is too large, subtle details of the image will be lost in the filtering process.

The algorithm consists of five separate identification and filtering steps. Multiresolution windows are evaluated to determine whether they contain point targets, a portion of a curvilinear feature, a homogeneous area, part of an edge, or isotropic data. After the maximum extent of the feature has been delineated, specialized filtering is applied for each feature type.

Algorithm

The following sections describe the five filtering steps identified by Touzi (2002).

Point Target Filtering: The Madsen point detector is applied. The ratio of the mean value of the inner window is compared with the outer window to determine whether a point target has been detected. If a point target is detected, the original value is preserved.

Curvilinear Filtering: If a pixel is not a point target, a three-strip ratio detector is applied to the subwindow in all four directions. The probability density function is used to compute the decision threshold associated with the user defined (CONTOUR) parameter. If a pixel is deemed part of a curvilinear feature, the next window size; that is, larger, is examined. The threshold of the larger window is calculated and the enlarged window becomes part of the curvilinear feature if the original direction is preserved, and the coefficient of variation does not change significantly. This process continues until one of these conditions no longer applies. A linear filter is applied to the maximum sized curvilinear feature.

Homogenous Area Filtering: If a subwindow is not a point target or part of a curvilinear feature, a test for homogeneity is applied. The local coefficient of variation is compared to the speckle coefficient of variation associated with the current window size and user supplied effective number of looks (NLOOK). If an area is deemed to be homogeneous, the expanded window is examined. This process continues until the maximum homogeneous area is found. The average of the pixels within the maximum homogenous area is written as the filter estimate.

Multiresolution Edge Detection and Filtering: If a subwindow is found not to be homogenous, the multiresolution ratio-edge detector is applied. The decision threshold of the detector is based on the user-supplied (EDGE) parameter. The edge detector compares the ratio of the means within various subwindows to determine whether a given pixel is part of an edge. If a pixel is classified as an edge; the expanded window is examined. If the expanded window threshold is maintained, the edge direction is preserved, and the coefficient of variation does not change significantly, the expanded window forms part of the edge. This process continues until one of the conditions is not met. An edge filter is applied to the maximum-sized edge.

Stationary Area Filtering: The remaining area is a nonhomogenous area that does not include a point target, curvilinear feature, or an edge. For each window expansion, the change in the coefficient of variation is examined. If the change is less than the maximum user-supplied (GRADIENT) parameter; the expansion continues. A structural multiresolution Frost filter is applied to the maximum isotropic area. The process continues until all of the remaining isotropic areas have been filtered.

References

Touzi R., 2002. "A review of speckle filtering in the context of estimation theory," IEEE Trans. On Geoscience and Remote Sensing, 40(11), pp. 2392-2404.

A. Lopes, E. Nezry, R. Touzi, and H.Laur, "Structure detection and statistical adaptive speckle filtering in SAR images," Int. J. Remote Sens., vol. 14, pp1735-1758, 1993.

A. Lopes, R. Touzi and E. Nezry, "Adaptive speckle filters and Scene heterogeneity", IEEE Transaction on Geoscience and Remote Sensing, Vol. 28, No. 6, pp. 992-1000, Nov. 1990.

R. Touzi, A. Lopes, and P. Bousquet, "A statistical and geometric edge detector for SAR images," IEEE Trans. Geosci. Remote Sensing, vol. 26, pp 764-773, Nov. 1988.

Zhenghao Shi and Ko B. Fung, 1994, "A Comparison of Digital Speckle Filters", Proceedings of IGARSS 94, August 8-12, 1994.

Touzi R., "A review of speckle filtering in the context of estimation theory", IEEE TGRS (Transaction on Geoscience and Remote Sensing), Vol. 40, No. 11, September 2002, pp. 2392-2404.

Touzi R., "A protocol for speckle filtering for SAR images", CEOS Proceedings, Toulouse, France, October 1999.