Abstract:
The orthorectification of the IKONOS image (9 February 2005) was done as part of the Heard Island AAS2939 project (Lucieer et al. in Jan - June 2008). This IKONOS image was orthorectified and co-registered to the IKONOS Jan. 2004 image (metadata ID: SIC_266_267_georectification). All processing was done in ENVI 4.4 (http://www.ittvis.com).
Quality
There were three IKONOS images covering areas of E-Heard, acquired on the 9 Feb 2005. There was one image which covered the entire eastern area, which was partly obscured by clouds. The other two cloud-free images together covered the same area. Therefore, the two cloud free images were used to form an image mosaic after pansharpening and orthorectification of the two individual cloud-free images. ... Pansharpening - Fuse or merge the high-resolution panchromatic band with the lower resolution multispectral bands to create a high-resolution multispectral image based on an image transform. - Panchromatic band has to overlap with multispectral bands in wavelengths. - Wavelet-based pansharpening is more correct from a spectral perspective than ENVIs pansharpening algorithms, such as HSV, Brovey, Gram-Schmidt, PCA, and CN. - Wavelet Fusion is an ENVI extension available from the ITTVIS codebank http://www.ittvis.com/UserCommunity/CodeLibrary.aspx - The original unprojected multispectral and panchromatic images were used for pansharpening. - The input images are transformed into wavelet space. Their wavelet coefficients are combined. Then an inverse wavelet transform is applied to the combined wavelet coefficients to create the fused image. - Fusion strength: the ratios used to blend together the wavelet coefficients of each image at each decomposition level. The settings used: - Image A: Multispectral image - Image B: PAN image - Map projection: Match Image B (PAN) - Fusion strength: 50:50 - Additional options: defaults - Output file name: include ps (pansharpened with discrete wavelet transform) - The resulting images had a 1m pixel size.
Orthorectification of the Pansharpened images
Orthorectification parameters in ENVI: - Orthorectification was done without GCPs as the match between the DEM and the QuickBird image was very good. Co-registration was applied after orthorectification to fine-tune the image rectification and to match the QuickBird image to the 2004 IKONOS image (metadata ID: "SIC_266_267_georectification". - RPC model based on the *_rpc.txt file from the panchromatic band - RADARSAT 2002 DEM was first reprojected in ArcGIS, so that it is in a geographical coordinate system (dem_gg.bsq). - The DEM resampling was set to Cubic convolution because it is upsampled to a higher resolution and cubic convolution results in a smoother interpolation. - The Geoid offset was set to the default of 0.0 as no accurate geoid offset was found on the Geoscience Australia website. - The projection was set to UTM, Zone 43 South with a WGS-84 Datum. - The pixel size was set to 1.0 m in both X and Y directions.
Mosaicking the two pansharpened and orthorectified images
- The cloud-free pansharpened and orthorectified 9 Feb 2005 IKONOS images were mosaicked to form a single image. The mosaicking was done based on image georeference information and the background value '0'(the black sections) was excluded in the mosaic. A fixed colour balance was set to equalise the colour between the two images. No feathering was applied to the mosaic.
Image-to-image registration of the mosaiced IKONOS 2005 image to the IKONOS 2004 image
- The 2005 IKONOS mosaic image was registered to the IKONOS 2004 image (metadata ID: "SIC_266_267_georectification") using image-to-image registration. 55 GCPs were selected before attempting to automatically generate tie points. - The more points selected manually, the more accurate the automatic tie points became. Many of these automatically generated tie points had very high RMS values, and upon closer observation it became clear that many of the points were located on unreliable locations, such as the ocean and glaciers. - In order to identify the most reliable GCPs it was necessary to stretch the image using the 'Enhance' tool to reveal particular areas in the image clearly. Different areas and land cover required varying enhancement methods. Rocks in the water and darker shadow areas were the areas which required more experimentation than other areas. During GCP selection the coastline was followed in the image, and GCPs selected on the outskirts of the island (e.g. rocks in the water). This was done to ensure that no land area would be excluded from the image during warping using the triangulation method. - The 145 GCPs chosen had an RMS Error value of 21.0. However, as we applied a triangulation (rubber sheeting) transformation, this RMS value has no significant meaning and does not reflect the accuracy of the GCP selection. - The mosaiced 2005 IKONOS image was warped using a triangulation method and nearest neighbour resampling. Nearest neighbour resampling ensured the pixel values retained their original spectral values.
Access Constraints
A PDF document providing further details is available for download from the URL given below.
Access to the image itself is limited owing to copyright issues. Contact the Australian Antarctic Data Centre for details and further information.
