There's More to An Image Than What Meets The Eye

By Dave Mohr

Characteristics of one-meter commercial imagery provides new benefits
The advent of one-meter commercial imagery such as Space Imaging's Ikonos and ImageSat's EROS 1 (soon to be followed by Quickbird 2 and Orbview4) offers an unprecedented opportunity for extracting timely and detailed information. Higher resolution, increased dynamic range, and betterrevisit rates over those of other satellite systems such as Landsat and SPOT provide the potential to extract not only better spatial information, but also better tonal information due to subtle contrast variations. This is indeed a huge, untapped information source. However, unless specific steps are taken during the exploitation and analysis phases to preserve the information benefits of one-meter 11-bit, satellite imagery, this potential will remain untapped-or even worse, destroyed-robbing the user of most of the improved information content of one-meter satellite imagery.

Issues that can affect information content
Resolution
Resolution is strongly determined by the aperture size of the satellite, the size of the detector elements, and the satellite's altitude. These are determined by the physical design of the satellite and cannot be altered significantly during the designed lifetime of the satellite. Nonetheless, a processing system designed to extract information from one-meter imagery will be of higher quality if it consistently and accurately preserves the spatial information that this higher resolution provides.

Geo-accuracy
True value in images lies not only in being able to identify features, but to determine precisely where those features are. The key here is accurate geo-registration. This must be built-in on the imagery supply side with accurate satellite positioning and pointing. The processing side must then be able to exploit that information. This is typically performed by supplying camera models or by use of Rational Polynomial Coefficients (RPCs). RPCs are mathematical expressions that approximate the complex mathematical relationships between the camera model and the image coordinates using low-order polynomials. These RPCs only apply within the limited region of the world where the image is centered. RPCs can provide a highly accurate substitute approach for image geo-registration in cases where the camera model is not available. The positional information for each pixel within the image must be preserved during all steps of the image-processing chain.

Dynamic range
Dynamic range determines how many shades of gray or levels of intensity can be distinguished in the image. While the human eye can distinguish around 100 shades of gray, there is considerably more information available in an 11-bit image. If the image has been captured with an instrument that can "see" more shades of gray than the eye, appropriate tools can process and display this information in a way that is useful to the human interpreter. The process is similar to taking closely spaced rings of a tree and "blowing them up" to see the fine detail in between. Think of it as magnification in "intensity space" as opposed to dimensional space that comes with the use of a magnifying glass.

Timeliness
Do small spatial events really happen more quickly than bigger ones? That's hard to say but, if they do, getting the information into the hands of the decision-maker becomes all that more important as the scale of the event approaches one meter, such as the exact determination of a flood boundary, the extent of agricultural pest outbreaks, neighborhood characteristics, or the scope of a military installation. Timeliness goes far beyond "just" collecting the information-a non-trivial task in itself. It includes the processing of raw data into a useful form (like turning a tree into 2x4s), delivery to the customer, enhancement, processing and interpretation of the image into useful information, and finally getting the information into the hands of the decision maker who, will take action based upon information derived from the image. Central tasks that can affect timeliness include:
• Collection
• Pre-processing
• Delivery
• Ingesting and Archiving
• Exploitation and Analysis
• Report generation and delivery.

Some factors that determine timeliness include:
• Degree of cloud cover over the area of interest
• Orbit of the satellite
• Efficiency of the imagery provider's production facility
• Broadband availability or speed of overnight delivery (for physical media)
• Ability to rapidly and efficiently extract information from the image
• Ability to rapidly convey the information to decision makers in an understandable format.

Steps to preserving information content-what can you do?
Isn't all the information content of the newly available 11-bit, high-resolution images automatically preserved? Unfortunately, this is often not the case. Most of the "legacy" systems, and software used for the exploitation of commercial satellite images, were developed for and optimized to handle lower-resolution, eight-bit images such as Landsat or SPOT. Using legacy systems on one-meter 11-bit images is almost like measuring with a micrometer, but cutting with a chainsaw. Much of the information in the image may be lost. When you pay premium prices for premium value, one-meter 11-bit imagery, it only makes sense to ensure that nothing in the processing chain destroys the value contained in the image.
      What can you do to preserve information content? Assuming that the image supplier has done everything practical to preserve the information contained in the images, up until the time the images are delivered, there are two key areas under your control that you can configure for optimal information extraction and timeliness: the exploitation and analysis stage, and the report-generation stage. Since report generation tends to be domain-specific, the remainder of this article will focus on the exploitation and analysis phase.

To compress or not to compress-maybe that's not even the question?
The first step in the exploitation process is reading the image data into the exploitation system, which may be as simple as a PC or as complex as an archival system combined with a spatial database engine and high-end workstations. Often the user will compress the image due to storage and bandwidth limitations. For many applications using lower resolution eight-bit images, this is a fine solution. However, as others have pointed out, compression is not advisable where very small image details or subtle variations of the data are important. Since two of the strong points of one-meter 11-bit satellite images are precisely small image details and subtle variations of the data, compression in this case may not preserve the very information for which the images were purchased. In addition, for users who want to exploit the information contained in multispectral data, compression may not be advisable because valuable information can be lost in the compression process.

Exploitation and timeliness: size matters!
Assuming for a moment that small image details and subtle variations of the data matter to the user, the challenge then becomes being able to manipulate the image data in a timely manner without using compression. This means working with single-frame images approaching one-gigabyte in size. Mosaiced images can be even bigger! This poses a challenge to many image exploitation packages with respect to timeliness. Screen-refresh or repainting rates may become agonizingly slow. Moving around in the image (roaming) and zooming in on details can become very frustrating. The most powerful or sophisticated processing algorithms can become practically useless if the user becomes frustrated with the first steps of the exploitation process of "simple" navigation. Most users want "seamless" panning, roaming and zooming functions that operate smoothly and without delay. The bottom line: not all image processing and GIS products can routinely handle large raster files. A "test drive" before you buy is warranted.

Throwing out the baby with the bath water: eight-bits vs. 11-bits
As already mentioned, the new commercial satellite data offers images consisting of up to 11 bits instead of eight-bits. What this means is that in an 11-bit image, there are 2048 levels of intensity available in the image instead of 256 levels in an eight-bit image-eight times the potential intensity fidelity of that in an eight-bit image. Even though the human eye cannot simultaneously distinguish more than about 100 levels of intensity, an image-processing system can. The trick is to center the human "window" of perception on the intensity point of interest. This window can be moved up or down the intensity scale to bring out all the fine details the 11 bits have to offer. By doing so, the user can preserve information in both the dark areas as well as the brighter areas of an image. An example of this is in extracting information out of shadows, whether from buildings or clouds. Figures 1 and 2 show an example of the ability to draw features out of shadows from an 11-bit image. Note the street detail available in Figure 2 that would most likely be lost if the image contained only eight-bits.
      Because many exploitation packages handle only eight-bit data, there is a growing demand for eight-bit one-meter satellite data. By "dumbing down" the images to 8 bits one is throwing away over 87 percent of the information content just to make the images "fit" the exploitation software.

Solutions are out there-Image Chain Analysis baseline
The U.S. government has amassed years of experience dealing with image quality issues associated with the digital processing of high- resolution, wide-dynamic-range "spy" satellites. Known as Image Chain Analysis (ICA), scientific studies and user experience resulted in specific rules for processing satellite information so that maximum information is preserved and available for human interpretation. These time-tested methods are used on a daily basis within the U.S. government to tap the huge potential of high-resolution satellite systems. Some of these lessons include how and in what order images should be zoomed, rotated, sharpened, and contrast-enhanced. The following are just a few of the critical issues addressed in the ICA processing guidelines.

Sharpening filters
Blurring in an image can sometimes be reduced with the use of a sharpening filter. Sharpening filters do this by enhancing edge details in an image. The sharpening function in an image-exploitation package can be built so the user can sharpen easily to whatever degree of sharpness he or she desires. When done properly, sharpening can enhance edge detail while preserving much of the overall tone of the image. This keeps the image looking like an image, not an abstract grouping of outlines and shapes. Improper sharpening can significantly boost noise without enhancing the visual content of the image. Sharpening can extract spatial information from an image that might be lost or overlooked, especially when the intensity levels across straight or curved borders are slightly varying. The use of sharpening filters is especially important in one-meter imagery, where many more edges and boundaries are present. Several examples show the detail that can be enhanced by using a sharpening function. Figure 3 uses a simple nearest-neighbor filter and a 4:8 zoom. Figure 4 is significantly clearer. It uses a Lagrange filter and a 4:8 zoom.

Tonal transfer curves
Through the use of appropriate tonal transfer curves (TTCs) a user can make an image lighter or darker, or can extract information from shadows or bright areas, without having to know anything about image statistics or redistribution of image histograms. Functionally this equates to changing the image contrast, allowing the user to move the limited "response window" of the human eye over the intensity levels of most interest. TTCs prescribed by the ICA are designed specifically for 11-bit processing and are designed to extract the maximum information from an image. Figures 5 and 6 contrast the kinds of enhancement that can be obtained with the appropriate TTC.

System calibration
TTCs may be combined with monitor calibration curves that correct each monitor's nonlinear response. Individual monitor and printer calibration ensures that the human perception ability is preserved to the maximum extent for each displayed scene.

Commercial availability and applications
Until very recently, full ICA-compatible software has not been available to the commercial marketplace. In fact, it is only with the advent of the new class of commercially available one-meter 11-bit satellite imagery that it made any sense to use image-processing software of this more robust class in the commercial world. Many one-meter imagery applications are possible in the commercial world, but one application we examine in particular is the use of one-meter imagery in conjunction with lower-resolution imagery and GIS databases by the Department of Agriculture of a northeastern state. In a following article we will present an example of how one particular image processing software package, RemoteView™ by Sensor Systems, is helping fuse one-meter imagery with other data sources as a way of improving the exploitation and analysis in terms of both quality and timeliness in an agricultural and environmental setting.

Back