It can sometimes be very difficult to produce accurate color tones when pan sharpening Landsat images. As discussed in previous articles, the Landsat panchromatic band sensor is not sensitive to visible blue light. As a result, pan sharpened images can exhibit unnatural color tones when the panchromatic band is substituted for the Intensity or other color space band in the pan sharpening process. PANCROMA provides several algorithms to address this problem and these have been previously discussed. However, it may be difficult to achieve desired color tones using even these sophisticated algorithms. It is possible to combine the PANCROMA ENHG algorithm with other pan sharpening methods to help overcome such problematic images. The ENHG method uses the NDVI vegetation index to enhance the Landsat band 2 (green band). Although this method can produce improved results and works well by itself, it can also be combined with other methods. This paper outlines a technique for combining the ENHG with AJISANE.
As an introduction, I have compared the results of several PANCROMA pan sharpening methods using Landsat data set p205r021_7t20000717_z30_nnXX.tif, acquired over Scotland. This image is characterized by lots of vegetation in proximity to the ocean and other water bodies. This combination can prove troublesome as histogram matching algorithms can be influenced by such combinations.
The first image to the right is the RGB color composite. This is the standard to which we will compare the pan sharpened results. The second image is a straightforward application of four file HSI processing without histogram matching. This pan sharpened image exhibits the pronounced bluish cast that is characteristic of the Landsat panchromatic band.
The third image has been pan sharpened using the HSI algorithm and also using nonlinear histogram matching. As you can see, the histogram matching has produced accurate color tones for some of the cultivated fields, but the ones with the darker vegetation tones are too blue.
PANCROMA offers several algorithms for improving the spectral fidelity of pan sharpened Landsat images. These include the ENHG, XIONG and AJISANE algorithms. Each of these use the information contained in a fifth band, the near infrared (NIR) in order to correct the spectral distortions. An example of the same image pan sharpened with the AJISANE method is shown to the right. The green tones are considerably closer to those in the RGB composite in this image.
It is also possible to combine the ENHG and the other algorithms. ENHG uses the NIR band to modify the green band, increasing the pixel values of vegetated areas and creating a new ENHG band file. This file can be used in place of the green band for pan sharpening. The ENHG file can be prepared manually or combined with the HSI method using the automatic utility. If prepared manually, it can be combined with the XIONG or AJISANE algorithms as well.
The final image shows the result of preprocessing the green band using the ENHG algorithm (using a factor of 0.07), then substituting the ENHG band for the green band in the AJISANE method. I also adjusted the water body tones using the 'Change Dark Pixels' utility on the Image Processing Data Input form using a cutoff of 42 and adding 20 to the blue channel. All three channels brightened by 20 and some contrast and sharpening applied. The green tones are a bit better and this image and the water looks OK. While not an exact match for the RGB image, is acceptably close.
This exercise has demonstrated how the spectra fidelity of pan sharpened Landsat images can be further improved by combining the PANCROMA ENHG and other pan sharpening methods, including the AJISANE, helping to achieve color tones in the pan sharpened image that approach those of the standard RGB composite.