Our former algorithm for the NOMAP and Quicklook products said that the pixel values above the brightest 0.1% and below the darkest 0.1% would be mapped to the extreme values, with a linear fit in between. For a majority of images, this was a good choice that showed excellent contrast but prevented too much saturation.

However, 0.1% (a thousandth) of a two Gigapixel image is still two million pixels. So if there were a particularly bright spot, like a rocky outcrop amid a field of dunes, or a particularly dark spot, like a cavern opening in a plain of boulders, then all the saturation would occur in that one area, washing it out completely, and lowering contrast everywhere else in the image. So the algorithm needed to be more adaptable. After a good deal of experimentation, the algorithm we settled on looks at the brightest and darkest pixels in a thumbnail version of the image, and uses those values for the extremes, instead of the values at 0.1%. We shrink a copy of each color band to 1/11th the original scale. Pixel values in the original below the darkest in the thumbnail are mapped to pure black, while pixel values above the highest are mapped to pure white. The stretched bands are then merged to make the color image. Hence, a bright or dark spot smaller than 1/11th x 1/11th of the image size will no longer dominate the stretch.

What this ultimately means is, our RDR Extras now show more detail in areas that would be completely washed out by the old algorithm.

For example, in this ‘cave’ image, the left is from the original RGB.NOMAP.JP2, while the right is the same product using the new algorithm. As you can see, previously you could not tell if there was a floor to the hole or if it sloped away to greater depths.

The new algorithm is used strictly for the JP2’s; the browse and thumb are already scaled down enough that it would not make a substantial difference with them. The new algorithm went into effect today; coincidentally we just started orbit 8000. Images with the new stretch will likely appear in upcoming weekly releases and we plan to reprocess everything with this change (and improved calibration) during the summer.

Here is a more specific description of our old and new stretch algorithms.

Old Algorithm

Determine the min and max DN value corresponding to the following cumulative
percents into the histogram.

For the full RDRs, the stretch is linear and the output is 10-bit and
preserves the special pixel values: 0 = NULL, 1 = Low Representation Saturation, 2 = Low Instrument Saturation , 1022 = High Instrument Saturation, 1023 = High Representation Saturation.

For everything else, the stretch is linear, 8-bits per band and each band
is done independently. Special pixels are not preserved.

New Algorithm

The JPG products (browse and thumb) are be made with the original algorithm, with the following exception.

• The Color JPGs use the same stretch as the IRB JPGs. (Otherwise, the black area around the rotated image change the stretch).

The JP2 products are stretched as described above, with the following changes:

• The min and max for the stretch are determined not from a cumulative percentage
  but from the min and max DN taken by reducing the image or band by a factor of

  11 * ( binning / binning of red band )

  . The reduction averages pixels.

Notes

1. Due to PDS format restrictions, the 3 bands of the color RDR (IR-RED-BG) are not
   stretched independently.
2. The ISIS applications hirdrgen and isis2raw are used for stretching. The latter for
   RDR quicklooks and NOMAPs.

The magic happens when we succeed at coregistering the IR and BG to the RED parts of the image to produce the center strip, false color images. More about this in an upcoming post. The maximum width of a color image is 4048 pixels. Some HiRISE images are 100,000 pixels long, which makes for a very long skinny image. These are affectionately dubbed “color noodles” by the HiPI (PI=Principle Investigator).

The image below illustrates where the color portion of the image is located. The zoomed in part of the same image just shows more clearly how the colors can offer more detailed geologic information than is available in the RED (black and white) image. For detailed information about the use of the color products and how they can be interpreted for scientific purposes, please refer to “Information for Scientific Users of HiRISE Color Products”

Is this what Mars really looks like? The images are not true color. The three color images taken by HiRISE are coregistered and stacked on top of each other. Then each color layer is assigned to red, blue or green, because those are the colors that are projected on your screen. So you can see how the word “color” becomes quite confusing. First, red is black and white. Then, we have all those I’s R’s and G’s and B’s! The color in HiRISE color products is really false color, because we are assigning a visible color to one that is invisible to human eyes. Also, there are only three wavelengths of light, not the full visible spectrum we are used to seeing. The RGB products are more similar to “natural” color. Even with HiRISE’s limited color capability, there is still an incredible amount of information gained by having the two extra wavelengths.

Why is there a garish green strip along the right side of the color image (left side in the nomap products)? You will notice this in some of the HiRISE color products. It will be apparent in the IRB, but not the RGB products. This is due to one half of the IR10 CCD having electronics issues during the earlier part of the mission. This problem was resolved for most cases, so that later images have both channels of IR10 — no green strip. Some of the earlier images were also able to be reprocessed to restore the missing IR information.

What is the difference between “RGB” and “IRB”? The RGB products are different than the IRB products in that the IR channel has been replaced by a “synthetic blue” layer that creates an image that is somewhat closer to natural color. In many of the images, the infrared band does not contribute a lot of information. The bands in this product have also been stretched to provide better contrast. In other words, the RGB images are more aesthetic. The IRB product is a science product. It contains the IR, RED and BG layers.

In the IAS viewer, you can turn the bands on and off to see what information each one contributes to a particular image. Use this button to switch from color to grayscale. This dialogue will also allow you to switch the color assigned to each band. The way the images are stacked in the HiRISE images goes like this:

Changing two bands to display the same color will show what kind of information is contributed by each band.

Below is a detail from PSP_004052_2045 showing the IRB color overlaid on the RED image. It is a beautiful example of how the color available in HiRISE images gives us new information that aids in interpreting the images. They are also just plain beautiful.