Pan-sharpening combines the colour information from a multi-spectral (MS) file, with the geometric information from the panchromatic band (Pan), at four times the resolution. A pan-sharpened product is available from Eurimage, but sometimes you may want to make your own. This technique should only be used with smaller, cropped images. It requires the MS (after resampling) and Pan images to be exactly the same size and perfectly registered, therefore it will not work properly with QuickBird Basic imagery except for very small areas.
Landsat 7 images can be pan-sharpened using the same technique.
To begin with, then, you need a Standard image. The MS file must be converted to 8-bit, combined in the desired band combination, cropped and stretched – the colour and contrast of the MS file should be close to what you want from the final product. Do not apply any sharpening or other filters. Save this file. The Pan image should be converted to 8-bit, without any other processing.
Convert the MS image to Lab (Image\Mode\Lab Color). Resize the image to 400%, using the Bicubic option (For Landsat use 200%). Note that this produces an image 16 times larger. You should save this file too, if you have the space.
Split Channels. (If at any time the Split or Merge Channels commands are not available make sure that you are dealing with an 8-bit image and that Layers have been flattened).
The three Lab Channels are labelled Lightness, a and b. The a and b channels carry the colour information, and the idea is to substitute the Pan data for the Lightness channel. For this, the Pan image needs to be exactly the same size and perfectly registered with it. One method of doing this is to copy the Lightness image on to the Pan image. Select the Move tool and move the copy so that it is in approximately the right place. Zoom in to about 400%, so that individual pixels are visible. The opacity of the layer can be changed with the numbers on the numeric keyboard. 1 gives 10% (almost transparent) and 0 gives 100%. Nudge the copy with the arrow keys while switching the opacity back and forth. When the image no longer appears to move when the opacity is changed, the images are registered (this is a lot easier to do than to describe). The registration will not look perfect, as the two images have such different resolutions.
Control-click (Mac: Command-click) on the layer containing the Lightness copy - this should be Layer 1 - to select its area and then crop the image, which should now be exactly the same size in pixels as the multi-spectral image. Delete the layer containing the Lightness copy. Save this file. Now you need to apply the Levels command to the Pan image to make it look as much like the Lightness channel as possible. Ignore areas of water, or green vegetation, which will not match (see below), and concentrate if possible on roads, buildings etc. Zoom in and out. The higher resolution of the Pan image can make it seem to have more contrast. You will need to proceed by trial and error, which is why it is important to save intermediate steps.
Now apply Merge Channels, in Lab mode, using the Pan image for the Lightness channel and the original a and b channels. Convert to RGB. This pan-sharpened image can now be resampled and sharpened if necessary.
In many cases, the colours of the pan-sharpened image will look very different from the original RGB image, especially if the original contains green vegetation. This is because the Pan data from QuickBird (and landsat 7) extends into the infrared, as a result, the colours of the pan-sharpened imagery may not look very natural.
 |
 |
|
3 2 1 image (resampled 400%) |
Lightness channel from 3 2 1 image |
|
 |
 |
|
Pan-sharpened. Note colour changes |
Stretched Pan image before merging. Infrared compnent is especially visible in pool and grass |
This guide covers the following procedures:
Converting 16-bit imagery to 8-bit
Note: Photoshop is not the ideal software for processing satellite imagery, which is normally processed with GIS software. This guide is for those who want to prepare images for printing or other display purposes. Using the imagery in this way implies a good deal of trial-and-error and subjective judgement. This guide assumes a basic working knowledge of Photoshop.
QuickBird imagery typically involves very large files. Photoshop has a limit of 30,000 pixels in each direction. You will need a powerful computer with plenty of disk space (at least four times the image size), and in any case it will usually be better to work on small parts of the imagery.
Catalogues: