Processing Satellite Images

• Pan merge
  • A common operation is to do a "pan merge", which combines low-resolution color band information with high-resolution pan-chromatic images.
  • This gives a color image with better resolution than would otherwise be possible.
  • There are a number of standard ways do this merge operation, the simplest being to use a HSV approach, where the Hue and Saturation come from the color bands, and the Value comes from the panchromatic band
  • Chris Hanson writes: "[this approach] works ok, but usually turns vegetation from a deep forest green to a minty pistachio green. This is because vegetation is very reflective in the IR band, which the pan band covers.  A good technique is to do a min() calculation between the pan band and the average of the three RGB bands, sampled with a smoothing technique that will allow you to sample between the two bands of different resolution.  This keeps the bright vegetation areas from blowing out the Value channel in the HSV merge.  Unfortunately, very few programs can perform interband math with images of different resolution."
  • Important academic papers: Smoothing Filter-Based Intensity Modulation (SFIM), Jian Guo Liu, 2000 (Paper 1, Paper 2) describes an approach which is better at preserving the spectral information of the original color input.
• Producing a "true color" image
  • This is difficult when the color bands do not include blue (as is the case with SPOT data)
  • One approach is to map (Red, Green, Near-Infrared) onto (Red, Blue, Green)
  • This somewhat works because the infrared usually indicates vegetation, which is usually green
  • This fails badly over regions like deserts, which lack vegetation and hence appear purple

Ways to Merge Satellite Bands into RGB Images

• this can be done using a variety of image software, a commercial service, or (with luck and many steps) a normal paint program
• PixelSense
  • an inexpensive ($50) Windows app which will merge raw LandSat7 imagery from the GLCF
  • preserves the GeoTIFF extents in the input files
  • attempts to guess a realistic mapping for the dynamic range in each band, although this can be thrown off by factors such as clouds - if the colors in the result don't appear as desired, you can always use a paint program to tune them after the merge
• Orfeo Toolbox / Monteverdi
  • Free, open-source French software which does the whole range of remote-sensing operations, including radiometric corrections, orthorectification, change detection, classification, etc.
  • It includes algorithms targeted at specific sensors, for optical images (SPOT, Quickbird, Worldview, Landsat, Ikonos), hyperspectral sensors (Hyperion) or SAR (TerraSarX, ERS, Palsar)
• tutorial on Terrainmap.com: LandSat HSL Composite from Panchromatic Band
  • shows how to merge bands with the program Paint Shop Pro ($100, or free for 30-day trial)
  • Paint Shop Pro apparently has a specific "Combine Channel" command
• tutorial by Abhijit Menon-Sen: Compositing your own LandSat-7 scenes
  • shows how to use GRASS to do a merge that is more sophisticated than simply compositing the visible color bands
• Purdue-LARS MultiSpec
  • a free (Windows and Mac) remote sensing analysis program
  • some simple instructions for how to make it do the merge operation
• MicroDEM (free, Windows only)
  • the author says "MicroDEM just loves that data (LandSat7 bands as TIF files), and can create a TIFF with three bands of your choice in a single file."
• PhotoShop
  • Steps:
    • load the 3 files containing the bands, ensure they are set to the "greyscale" image type
    • then use the "Merge Channels" command which merges them into an RGB image
    • then use "Image: Adjust Levels" to narrow the range of each R,G,B until the image looks good
    • a drawback to using PhotoShop is that it does not preserve the coordinate information GeoTIFF files, so it is hard to do geographic operations, like cropping based on coordinates, after processing with PhotoShop
    • one solution is to do any cropping operations beforehand, saving the PhotoShop merge as the last step
  • these steps are illustrated nicely in Scott Cherba's tutorial Combining LandSat Bands in Photoshop
  • reportedly, the Digital ROC plugin ($50, Windows/Mac) makes the process of adjusting RGB levels easier, although i found i got similar results with more control using PhotoShop's built-in Adjust Levels - your experience may be different, so the plugin may be worth trying
• using a paint program like Paint Shop Pro or PhotoShop allows for other operations that are almost always necessary, including selecting specific areas of the image to fix hue/contrast, and editing out clouds, best done with the "rubber stamp"/"clone" tool

Cropping Large Images

• working with huge images can be difficult, especially when your area of interest is much smaller than your source files
• for cropping each band before the merge, Global Mapper is a good option ($180, Windows, free version with less features), since you can directly type in the extents of your area of interest, though you need to do so 3 times, once for each band
• for the command-line oriented, GDAL can be used
  use gdal_translate with the [-projwin ulx uly lrx lry] option