Landsat Image Registration

Image Registration - In order to use the PANCROMA Landsat gap-filling and cloud mask utilities you must provide two image sets. The first set, which I call the Reference set, typically has some feature that needs to be corrected using data from a second data set, that I call the Adjust set. For example, the Reference set may have gaps, that is, missing data as a result of the failure of the Landsat Scan Line Corrector (SLC) in 2003. PANCROMA gap filling utilities use the data in an image set collected prior to the failure of the SLC to fill in or to compute the image information that should go into the gaps. Likewise the Reference image may not suffer SLC-off gaps but may have considerable cloud cover. In this case PANCROMA cloud mask utilities can detect and replace the clouds with black pixels, so that they can be subsequently filled in from a suitable Adjust data set without clouds.

In either case, the Reference and the Adjust data sets must be perfectly registered or matched so that the data can be transferred from the Adjust set to the Reference set seamlessly, producing a natural looking image. PANCROMA provides a collection of utilities to produce this registration. These utilities often contain the following elements:

• Compute Maximum Extents - this utility determines the maximum overlap between the Reference and Adjust data sets and specifies this overlap by computing the corner coordinates that define the overlap

• Image Rescale - Sometimes the Reference and Adjust images do not have the same scale factor. Landsat band images for example can have either 30m or 28.5m resolution. In this case the images must be rescaled so that the two match, i.e. the 28.5m image must be rescaled to 30m resolution

• Image Subsetting - Next the two images must be cropped according to the Maximum Extents corner coordinates so that both images have the same latitude and longitude boundaries

• Row and Column Count Rationalization - The images get a final check to correct any off-by-one row or column discrepancies resulting from rounding errors when recomputing row and column counts during any rescaling and when mapping the corner latitude and longitude values to their corresponding row and column values.

Formerly, this process was done using four separate, semi-manual utilities. A new PANCROMA utility conducts these operations automatically, saving time and reducing the probability of making an error in the process. In order to use the utility, you must start with two three-file data sets, for example a Landsat band1, band2 and band3 Reference file and a band1, band2 and band3 Adjust file. There must be at least some common area shared between the two data sets. The band files must also be in GeoTiff format. To get started, open three band files (either the Reference or Adjust group) by selecting 'File' | 'Open' and opening three files from the same data set in series. Continue opening three more files, for example the files from the Adjust data set if you opened the Reference set first. You should now have six band files opened. Now select 'Band Combination' | 'Subset Images' | 'Subset Six Bands'. When the subset data entry box appears, just click on 'Enter'. PANCROMA does the rest after that. If all goes well, you should see your six registered images appear after a bit of computation. You can save all six files by selecting 'File' | 'Save' and typing in a base file name. PANCROMA will add a numerical suffix to the base file name, numbering the output files in the same order that you input them. You are now ready to gap fill, cloud mask, or otherwise process the registered images.

The example files to the right show how the utility works. The first image is Landsat file L71001076_07619991115_B30.TIF, Row 76 Path 1 acquired over Anotfagasta, Chile. The second image is Landsat p001r076_7t20000525_z19_nn30.tif, Row 76 Path 1 of the same area. The first image has 6991 rows and 7851 columns and is a 30m per pixel image. The second image has 7641 rows and 8704 columns and is a 28.5m per pixel image.

The next two images are the registered versions of the two images, having the same corner coordinates, same scale, and same row and column counts. When superimposed one over the other the features match very closely. The details to the right show the boundary area between the two image edges, one fully opaque and one semi-transparent. (The image edges do not match perfectly because PANCROMA matches the collared area corners, not the exposed image within). As you can see, the road and dry watercourse features align almost perfectly along the western edge of the image.

At the present time, PANCROMA can only match three-file data sets using the automatic utility, and data must be in GeoTiff format. Subsetting is to Maximum Common Extents only. I will probably expand the utility to handle BMP and JPEG format and also four-file data sets for pan sharpening. However I believe that three-file sets represent the majority of maximum.