PANSHARP

Description

PANSHARP creates high-resolution color images by fusing black-and-white panchromatic and multispectral color images. This image fusion technique is known as pansharpening and supports 8-bit, 16-bit, or 32-bit real data types. PANSHARP can fuse images obtained from the same sensor or from different sensors. Although other image fusion techniques such as Intensity, Hue, and Saturation (IHS) and ImageLock Data Fusion can also be used, PANSHARP produces superior sharpening results while preserving the spectral characteristics of the original images.

Parameters

Name	Type	Length	Value range
Input: Input multispectral channel	Raster port	0 - 1024
InputRef: Input reference image channels	Raster port	0 - 1024
InputPan: Input panchromatic image channel	Raster port	0 - 1
Source background options	String	0 -	Default: File Metadata
Source background values	Integer	0 - 1024	Default: 0
Enhanced pansharpening	String	0 - 1	YES | NO Default: YES
Resample Method	String	0 - 1	CUBIC | BILIN Default: CUBIC
Output: Output channel	Raster port	0 - 1024
Report	String	0 - 192	See parameter description

Parameter descriptions

Input: Input multispectral channel

Specifies the multispectral channel(s) of the input image file. These channels are fused with the high-resolution panchromatic image data. Duplicate channels are not allowed.

Any multispectral channel can be used, including those that are outside of the range specified by DBIC_REF (InputRef). Using the channels specified by DBIC_REF (InputRef) yields more accurate results.

InputRef: Input reference image channels

Specifies the reference image channel(s) in the input file. These channels and the channels of the panchromatic image span the same range of frequency (wavelength) response. Duplicate channels are not allowed.

The use of DBIC_REF (InputRef) channels varies from sensor to sensor. For more information, refer to the Details section.

If this parameter is not specified, PANSHARP determines the appropriate reference bands based on the available wavelength information for both FILI_PAN (InputPan) and DBIC (Input). PANSHARP generates a warning if it cannot find any metadata and use all the bands in the input file.

PANSHARP will generate an error if the wavelength is present and there is no overlap.

InputPan: Input panchromatic image channel

Specifies the channel that contains the high-resolution panchromatic image data from FILI_PAN. Only one channel must be specified. By default channel 1 is used.

Source background options

Specifies, potentially with Source Background Values, which pixels in the source images are to be considered background (NoData) pixels. The same rule is applied to both of the input images independently. In general, if a pixel is considered NoData, the application handles the pixel in a special manner.

See the Source Background Values parameter for specific examples.

Source background values

Enhanced pansharpening

Specifies whether to generate a refined pansharpened image. The default value is YES.

Resample Method

Specifies the resampling method to use to determine the output pixel values when resampling the input low-resolution data to the higher resolution panchromatic data.

Supported values are:

• CUBIC: cubic convolution; this is the default value.
• BILIN: bilinear interpolation.

Bilinear interpolation is recommended if further analysis (such as classification) using the output is expected, as it will keep the output values closer to the original MS pixel values.

Output: Output channel

Optionally specifies the channel(s) to receive the output pansharpened input channels. If this parameter is not specified, the first N channels are used, where N is the number of selected input channels.

Duplicate channels are not allowed.

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

PANSHARP applies the automatic image fusion algorithm to increase the resolution of multispectral (color) image data by using a high resolution black-and-white (panchromatic) image.

The power of PANSHARP lies in the simplicity of its algorithm and its versatility. It works with any image data type (8-bit unsigned, 16-bit signed/unsigned, 32-bit floating point) and is computationally efficient.

For best results, the input reference image channels must be selected in such a way that the multispectral bands cover as closely as possible the frequency range of the high-resolution panchromatic image.

The following list shows the reference bands for some well-known satellite sensors:

Landsat 7 (ETM+):     Green:2, Red:3, Near IR:4
SPOT 1, 2, 3 (HRV):   Green:1, Red:2
SPOT 5 (HRG):         Green:1, Red:2
IRS 1C, 1D:           Green:1, Red:2
IKONOS:               Blue:1, Green:2, Red:3, Near IR:4
QUICKBIRD:            Blue:1, Green:2, Red:3, Near IR:4
Worldview-2 (4-band): Blue:1, Green:2, Red:3       
Worldview-2 (8-band): Blue:2, Green:3, Yellow:4, Red:5, Red Edge:6

The channel number given in the table above is the standard ordering on the sensor and may differ from the order in an actual data file.

The order in which the reference input channels are specified is not important.

The better the co-registration of the input panchromatic and multispectral image data is, the better the results from this function will be. If a geometric correction pre-processing step can improve this co-registration, then it should be considered.

Atmospheric differences between the times at which the panchromatic and multispectral images are acquired will reduce the quality of the results from this function. Images that have been acquired simultaneously should be used if possible. Landsat 7, IKONOS, Quickbird and the Spot series of satellite-borne sensors deliver simultaneously acquired panchromatic and multispectral image data.

It is recommended that the ratio of ground sample distances between the multispectral and panchromatic images not exceed 5:1. For example, IKONOS data with 4m MS and 1m Pan have a ratio of 4:1, which is acceptable. However, Landsat 7 30m multispectral images have 50 times the ground sample distance of Quickbird 0.61m panchromatic images, and an attempt to sharpen the former with the latter would produce poor results.

The mean, standard deviation, and histogram shape for each multispectral image is typically only slightly modified by sharpening with this function, so long as the ground sample distance ratio guideline described above is followed and the set of reference images closely cover the same wavelength range as the panchromatic image.

If a pixel is NoData in either of the input images, it will be NoData in the output. Regardless of the value of the input NoData, the output NoData value will always be 0.

Algorithm

Based on thorough studies and analyses of existing fusion algorithms and fusion effects, a new automatic fusion approach has been developed. This new approach solved the two major problems: color distortion and operator/data dependency.

An approach based on least squares was developed to best approximate the gray value relationship between the original multispectral, panchromatic and the fused images to achieve a best color representation.

Statistical approaches were developed to realize a standardized and automated fusion process.

How to get the best results

A number of image fusion algorithms have been developed targeting different image sensors or applications. Typical problems found with existing algorithms include: (1) color distortion and (2) operator and data set dependency. Although the image fusion technique implemented in PANSHARP diminishes greatly those deficiencies, no algorithm is known to produce "perfect" results. The pansharpened multispectral images will inherit the level of spatial details from the panchromatic image data but won't benefit from sufficient color information to guarantee "true" high resolution multispectral images. Thus, spatial features of sizes smaller than the multispectral sensor resolution may be assigned "fake" color although visually the image quality appears to be excellent.

Another very important aspect of image fusion algorithms is ensuring very good co-registration of the input panchromatic and multispectral image data. This problem becomes more pronounced with high resolution sensors as IKONOS or Quickbird. A single pixel offset between the high resolution panchromatic image data and the low resolution multispectral image data results in color shifts clearly visible in the pansharpened imagery. Thus, a pre-processing step such as geometric correction or orthorectification, may be required before pansharpening the multispectral image data.

For best results, it is highly desirable that the high resolution panchromatic image data and the low resolution multispectral image data be acquired simultaneously. This will ensure consistency of the spatial and spectral features of the scene. Landsat 7, IKONOS, Quickbird and the Spot series of satellites offer simultaneously acquired imagery.

Due to the lack of sufficient color information in low resolution multispectral imagery, it is recommended that the ratio of resolutions between the images do not exceed 5:1. For example, IKONOS with 4m MS and 1m Pan has a ratio of 4:1 which is acceptable. However, mixing Landsat 7 30m MS with Quickbird 0.61m Pan would have a ratio of about 50:1 and the resulting fused image would have poor color quality.

PANSHARP produces best results for multispectral image channels whose wavelengths lie within the frequency range of the panchromatic image channel. Multispectral channels outside the wavelength range of the high resolution panchromatic image channel will still look good but may have reduced physical meaning.

Spectral characteristic preservation

The PANSHARP algorithm attempts to preserve spectral characteristics (that is, the color of the image). The mean, standard deviation and histogram shape for each channel are approximately preserved. Significant deviations from the original histogram shape can occur, however, if the resolutions of the multispectral and panchromatic imagery differ greatly or if data from different sensors is used together.

For more information and comparative results, see the References section.

Acknowledgements

The automatic image fusion algorithm was developed by Dr. Yun Zhang from the University of New Brunswick.

Dr. Yun Zhang
Department of Geodesy and Geomatics Engineering
University of New Brunswick
Fredericton, New Brunswick
Canada E3B 5A3

References

The following references describe the algorithm that is implemented by this function.

Zhang, Yun. (2002). "Problems in the fusion of commercial high-resolution satellite as well as Landsat 7 images and initial solutions". In ISPRS, Vol. 34, Part 4, GeoSpatial Theory, Processing and Applications, Ottawa, Canada.

Zhang, Yun. (June 24-28, 2002). "A new automatic approach for effectively fusing Landsat 7 as well as IKONOS images". IEEE/IGARSS'02, Toronto, Canada.