Pan sharpening natural color island terrains can present some challenges. The previous article in this series discussed the spectral distortions inherent in the pan sharpening process and some of the utilities that programs like PANCROMA provide for correcting them. Island images or coastal areas do not respond as effectively to histogram matching and corrective algorithms like XIONG. (ENHG will always work as intended but will not necessarily produce the most pleasing green color tones.) This article will examine a difficult island image and will explore several approaches to producing an acceptable pan sharpened outcome
The target image is a Landsat file set for French Polynesia. The first image on the right shows the reference RGB rendering from bands 3, 2 and 1. Performing a standard four-band pan sharpening run without histogram matching or any other preprocessing utility produces the usual problems: dark image, nearly black ocean and pronounced blue vegetation tones. These are apparent in the second image to the right.
The next two images apply two different preprocessing techniques that each produce similar outcomes. The first image brightens the panchromatic input image by increasing the PANCROMA Pan File Level slider bar. The second image applies a linear histogram match. Both of these have the effect of translating the histogram of the panchromatic image to the right, i.e. uniformly brightening all of the pixels by a constant factor. The difference is that linear histogram matching will always align the distribution averages of the input and reference histograms while Pan File Level adjustment allows you to position the output histogram mean wherever you want. These have the effect of improving the ocean color but do virtually nothing for the unnatural vegetation tones.
The next image shows the effect of a standard four band pan sharpening run plus nonlinear histogram matching. The ocean and vegetation tones are better, but now some very unattractive blue highlights have been introduced. These are a result of the histogram matching algorithm used by PANCROMA and in fact all pan sharpening applications. Such algorithms create a discrete transformation function by matching the cumulative histograms of the input and reference image. In most cases this transformation function, when applied to the panchromatic grayscale band will yield an adjusted panchromatic image that more closely matches the computed intensity band without any side effects.
Occasionally however, the transform function will produce artifacts. One of the more common artifact is unnatural light blue image highlights as in this example. This is most common and in fact can be pronounced in images where a lot of green foliage and ocean are both present. The next image shows an example of these artifacts. When histogram matching plus XIONG processing are performed, the base tones are further improved but the blue highlights are still visible.
The next run shows the result of a standard four band pan sharpening run with nonlinear histogram matching and with the PANCROMA 'Remove Highlights' check box checked. 'Remove Highlights' invokes a preprocessing algorithm that attempts to adjust the transformation function to alter the channels that produce the blue highlights, thus reducing them. The algorithm can be enabled by checking the PANCROMA 'Remove Highlights' check box in the 'Histogram' data group within the Image Processing Data Input box. When you do so, two Combo Boxes will become visible. These are used to set the lower and upper limits of the transformation function levels to which the correction will be applied. These levels are defined in terms of fractions and multiples of sigma, the standard deviation of the output histogram. The lower level is equal to the histogram mean plus sigma plus the standard deviation divided by the lower level integer. The upper level is defined by the histogram mean plus sigma multiplied by the upper level multiple. (The algorithm will decrease the value of all channels within its span in its attempt to correct the highlights, so it is important to exclude bright cloud, snow and surf pixels on the upper end and vegetation and ocean pixels on the lower end.)
lowerLevel = xBar + (sigma/lower Integer)
upperLevel = xBar + (sigma * upperInteger)
When the 'Remove Highlights' check box is selected, a Combo Box is displayed that allows the user to select the multiplier that will be applied to the standard deviation of the histogram in order to establish an upper limit of the processing range. (The algorithm will decrease the value of all channels within its span in its attempt to correct the highlights, so it is important to exclude,bright cloud, snow and surf pixels on the upper end and vegetation and ocean pixels on the lower end.) This image required a divisor of '6' and a multiplier of '8' to produce the next image in the series. You can see that the blue highlights have been considerably reduced
The final image shows the result of applying the correction along with five file XIONG processing plus histogram matching. This image, although it has minor imperfections, matches the color tones of the RGB reference image fairly closely.
Although processing island images can be a challenge, PANCROMA preprocessing algorithms help to achieve consistently good results.