Pan Sharpen Landsat SLC-Off Gap Filled Images

This article describes the process for creating a natural color Landsat image by first gap filling and then pan sharpening the gap filled input files. The described procedure will create four matched multispectral bands and a matched panchromatic band for each of the two file sets: the Reference set (the one with gaps) and the Adjust set (the one without the gaps that will be used to correct the Reference set.) I used a pair of Landsat scenes from Row 91 and Path 76 in southern New Zealand for this example.

The first step is to assemble your five Reference images and five Adjust images: a set of four multispectral bands and a panchromatic band for each. The multispectral bands should include the blue, green, and red and near infrared (NIR) Landsat bands. You will need an exactly matched set of files. PANCROMA provides the necessary subsetting and resizing utilities to produce these. The process will occur in four steps: subsetting, resizing, gap filling, and finally pan sharpening.

To start, select 'File' | 'Open' and open the first file set (you can use either the Reference or the Adjust set). You must open them in the precise order: blue (band1), green (band2), red (band3) panchromatic (band8) and NIR (band4) in your Landsat set. Now select 'Band Combination' | 'Subset Images' | 'Subset Five Bands'. After a bit of processing, the Subset Data Entry box will be presented. Subset the five bands simultaneously by any of the usual means for corner selection: rubber band box, pixel coordinates or latitude and longitude specification. Save the images by selecting 'File' | 'Save Subset Images' | 'GeoTiff' (or whatever format you like) and supplying a base file name like 'reference'. Select 'Close Graphics Window and Reset'. (Note: the images will not become visible until the 'Save' operation.

Now open the second set of five band files as described above. The Subset Data Entry box will again become visible. Notice that the pixel coordinates and the latitude and longitude values from subsetting the first set of images are still in the text boxes. This will allow you to select the same geographical area in the second set that you selected in the first set. Since the two Landsat scenes will differ somewhat in size and in area covered, you cannot select the second set by pixel coordinate value. Instead, select the 'Select by Coordinates radio button and then select 'Enter'. Note that if you select a subset area in the second set that is outside of the geographic extents of the first set (something that is easy to do if you select an area near the image edges) then you will not get a matching subset. This is because PANCROMA will truncate the selection in order to prevent inputting illegal coordinates. PANCROMA has a feature to automatically select the full common areas of both scenes. See the Instruction Manual for more information on this feature. When you have successfully subsetted the second set, you can save it as before by selecting 'File' | 'Save Subset Images' | 'GeoTiff and supplying a base file name like 'adjust'. Select 'Close Graphics Window and Reset'.

You now have two sets of subsetted images that have the same corner latitude and longitude coordinates but probably have different row and column sizes. Since the subsequent operations require consistent row and column values, you must resize the larger file set down to match the smaller one. To do this select 'File' | 'Open' and alternatively hover over your Reference and Adjust band files. The size of the tiff files will be reported at the tool tip. Note the row and column dimensions of the smaller set and open the larger four multispectral bands. Note: DO NOT open the panchromatic band at this time. Select 'Pre Process' | 'Resize Images' | 'Resize Four Images'. When the images and the Resize Image Data Entry box becomes visible, enter the row and column dimensions of the smaller file set and click 'OK'. After the images are resized, select 'File' | 'Save Subset Images' | 'GeoTiff and supply a base file name like 'resizedAdjust'. Select 'Close Graphics Window and Reset'.

The final step before gap filling is to separately resize the larger of the two panchromatic bands to match the smaller. Again select 'File' | 'Open' and alternatively hover over your Reference and Adjust panchromatic band files to determine and record the row and column count of the smaller set. (Note: the row and column sizes for the two panchromatic bands must be consistent with both sets of multispectral bands, i.e. exactly double AND match each other. If this is not the case additional resizing may be necessary.) Open the larger of the two files and select 'Pre Process' | 'Resize Images' | 'Resize One Image'. When the images and the Resize Image Data Entry box becomes visible, enter the row and column dimensions of the smaller file set and click 'OK'. After the images are resized, select 'File' | 'Save Grayscale Image' | 'GeoTiff and supply an appropriate file name like 'resizedReferneceBand4'. Select 'Close Graphics Window and Reset'. Select 'Close Graphics Window and Reset'.

You should now have two matched sets of five file sets each, one with gaps (Reference set) and one without (Adjust set). The next step is to fill the gaps. This is done exactly as described in previous tutorials and in the Instruction Manual. Open the files pair-wise, first a Reference band (one with gaps) and then the corresponding Adjust band (one without gaps.) For example, start with referenceBand1 and resizedAdjustBand1 (or whatever you naming convention was.) Select 'Gap Fill' | 'Gap Fill Hayes Interpolation Method' or 'Gap Fill' | 'Gap Fill Transfer Method' If you choose the Hayes method, a Gap Fill Data Form will appear. You can then select the algorithm parameters: 'Search Extents', the 'Gap Threshold' and the 'Comparison Radius'. Select 'OK'. After the file is processed, save the gap filled band by selecting 'File' | 'Save Grayscale Image' | 'GeoTiff and supply an appropriate file name like 'gapBand1'. Repeat the process four more times for the band 2, 3, 4 and 8 (panchromatic) file pairs. When you are done, select 'Close Graphics Window and Reset'.

Now you can pan sharpen you gap filled images as usual using the HSI or Brovey methods as described in previous articles and the Instruction Manual. I posted two sets of gap filled images from different parts of the Landsat Scene. I chose a challenging image that contains lots of ocean, clouds, snow and mountains. My results are shown to the right. The first image is an RGB color composite of the Reference bands showing the extent of the gaps in the original image. The second image is an RGB color composite of the gap-filled bands. The third image is a pan sharpened image produced using XIONG processing and the NIR band. There is a second set of a different part of the scene shown below the first set.

Both the image detail and the color fidelity are acceptable to me. The only major fault is the presence of some pixel artifacts toward the right edge of the second set of gap filled images. These are the result of snow cover on the mountain peaks in the Adjust image that were not present in the Reference image. This can be seen in the final grayscale Adjust image for band1. If the Hayes algorithm is not able to compute a suitable match, it simply transfers pixels from the Adjust image to the Reference image. This occurred for a number of saturated pixels in the snowfields. Since each of the four bands is matched separately, sometimes a pixel was carried over only in the red band, for example, resulting in a bright red pixel in the color image. The effect was somewhat reduced by the pan sharpening algorithm, but was not entirely eliminated. A better selection of Landsat scenes or add ional image processing is necessary to solve this problem.

Gap filling is very necessary and beneficial for SLC-Off Landsat RGB color composites. This article has shown how it can also be applied to pan sharpened images as well.