As described in a previous post, HiRISE color images are made by combining images in three different wavelengths of light, infrared (IR), red (RED) and blue-green (BG). The incoming light from the surface of Mars is separated by a filter into these three parts of the spectrum. The detectors that receive those wavelengths of light then build up the three separate images of the same place on the surface. The IR and BG detectors are above and below the RED detectors in the HiRISE focal plane, so they are imaging the same place, but at slightly different times. In order to create the color products, the three separate images have to be stacked one on top of the other. Lining up these images perfectly with each other is called coregistration.

This process seems simple in concept, but in practice it is quite complicated. There are three factors to account for:

• Relative timing
• Pixel binning
• Spacecraft jitter

First of all, HiRISE nearly always uses different resolutions for each color. For instance, RED might be at a scale of 25 cm/pixel (bin1) while IR and BG are at 1 meter/pixel (bin4). This “binning” minimizes the amount of data that has to be sent back to Earth, which is the most important constraint that HiRISE needs to deal with. Another reason for binning color is to improve the signal-to-noise ratio (SNR). This means that you can get a better signal by combining pixels at the expense of spatial resolution. In order to get the binned IR and BG images to line up properly, they must be enlarged to match the dimensions of the RED image. For example: RED is bin1, IR and BG are bin2. The RED image will be 2000 pixels wide by 40,000 pixels long (for example), the corresponding IR and BG images will be 1000 pixels wide by 20,000 pixels long. So the first step is to make the dimensions of all three images match.

Now the relative timing is easy to take care of. The start time of all images is a known quantity, and does not change from image to image. We know exactly when the BG detector starts imaging, followed by the RED detector, followed by the IR detector. So the beginning of each image is offset by a fixed amount. Once the images are shifted by this fixed offset (accounting for binning), they will be approximately lined up.

This brings us to the third factor in coregistering images — spacecraft jitter. Because HiRISE is imaging at such a high spatial resolution and at great speed, tiny motions of the MRO spacecraft cause slight variations in where the surface features appear in each of the three color detectors. Imagine that HiRISE is taking a picture of a 1m sized boulder on Mars. If the rock shows up in line 100 of the RED image, and we have already accounted for the relative offsets of the detectors and for binning, then the rock should also show up in line 100 of the BG image. But say we look and it is actually in line 99. Now when we try to stack the two images, the objects in them won’t line up exactly. Our color processing software corrects for this by holding the RED image fixed, and adjusting the corresponding BG and IR images to match it precisely. This is not a perfect process, but most of the time it works extremely well.

Producing the color HiRISE products is not a trivial process. But it is to a point where the processing is automated so new data is released without delay. Enjoy the colorful view!

While spending a little time working on various scalebar utilities, I thought I’d take a look at the distribution of image sizes. We’re at around one thousand images (TRA + PSP) of Mars. This is over a half million megapixels in total, not counting calibration images.

The first plot is a histogram of total pixels. Most frequently, images are under a quarter gigapixel, but there are quite a few between 750 and 1000 megapixels. And it has been uncommon to take images larger than a gigapixel.

The next plot is a histogram with 2-D bins, showing the frequency of width and height (in pixels).
The more popular imaging modes are colored brighter. So the most common dimensions are around 20,000 by 40,000 (the yellow box). Most images are the full width (20,048 pixels), followed by half width, etc (the columns). This is simply a function of the binning (subsampling) modes, but can change due to missing or unusable channels. However, there is a good deal of variability in the heights, which corresponds to the exposure duration (and also the binning). At least one image had over 120,000 lines!

These plots were generated with ploticus, a free-software plotting utility.