| EOM Airborne: Minimizing Subjective Image Interpretation in Digital Orthophotography
By Mark Klimiuk The production, acceptance and use of Digital Orthophoto Imagery (DOI) has improved significantly over the past few years. Prior to deciding whether DOI will be a component of a GIS, the functionality and use of DOI must be determined. Once the decision is made to utilize DOI technology, an understanding of the product and its development process is necessary to define production techniques, acceptance criteria and expectations. Those requesting DOI with little information or experience have extremely high expectations for the product, and inevitably may become disillusioned if all expectations aren't met. If DOI is to become understood and accepted by the GIS community, defining project specifications at the outset is critical to satisfying expectations at delivery.
DOI can be separated into two distinct categories for evaluation, photogrammetry and imagery. The photogrammetric process includes Aerial Triangulation (A.T.) and Digital Terrain Model (DTM) collection. There are standards for mapping accuracy that can be easily measured on a DOI to satisfy photogrammetric acceptance. The aspect of DOI image quality is difficult and very subjective - beauty is in the eyes of the beholder. It is very difficult to select a control image during the pilot phase of the project and match it exactly throughout the project. The terrain and content of the imagery may vary significantly in a project area. It is very easy to show a digital orthophoto to a group of people and have various opinions formulated about its quality and interpretability. The only way to define acceptance criteria for DOI is to first understand, categorize, and quantify subjective issues related to image quality. Aerial Imagery
Throughout the aerial photography production and handling process there exists the unfortunate problem of particulate matter such as dust or lint coming in contact with the original film or diapositive. When the aerial photography is viewed at high resolution, these sometimes microscopic particles can appear to be quite large. The occurrence of these particles will be more noticeable in small scale imagery because of the high magnification usually involved during scanning. Another problem on the photograph are scratches. Scratches can be introduced at any time during handling of the photographic materials. The majority of scratches are created during the aerial photography and thus are in the imagery from the beginning. The worst type of scratches are introduced while the film is in the camera, and are called pressure scratches. These scratches can be on either the base or emulsion side of the film, or both, and are caused by the film contacting the platen or other surfaces inside the camera during frame advance. The scratches found on the base side of the film negative can only be detected by projecting a point light source at an oblique angle to the film. When a diapositive is created, the scratches will appear as thin white lines. A scratch that is not easily detectable in the original aerial film will now show in a reproduction. As with particulate matter, scratches will be more prevalent as scanning magnification increases. Particulate matter and scratches can all be corrected in the digital image.
The film required as input to DOI must be selected, exposed, developed and handled differently from other aerial film used for non-DOI purposes. The time of day, year and geographic features determine radiometric properties of the photography. The geographic area of photography, whether coastal, desert or urban will further determine radiometric properties. Special filters may be required to photograph through certain atmospheric conditions such as haze. The natural reflectance of materials both man-made and natural cause not only interpretability problems but in extreme cases, flare-outs. Steel roofs, water bodies and snow all exhibit reflectance problems. Smaller scale photography exhibits a phenomenon known as "hot spots" in an image. A "hot spot" is a portion of an image that is considerably brighter than the rest. It is safe to assume that most digital or analog dodging techniques can remedy these problems during diapositive creation.
Sun angle determines the magnitude of shadows that are present. Shadows are a natural occurrence that cause interpretability problems in DOI. Expectations about shadow detail should be defined prior to photography as this is the best place to begin. Hardware Limitations
An 8-bit gray scale DOI or 24-bit color DOI is comprised of picture elements or pixels. A gray scale DOI has 256 levels of gray, 0 (black) to 255 (white). It must be the intent during scanning to capture as much detail as possible in the analog photographic image. Most scanning production systems today use a charge-coupled device (CCD) array to digitize the aerial photograph. CCD arrays have their own limitations, one being that they only linearly function in density ranges of around 1.0. Photography that is characterized by a density range larger than this may not be accurately captured. Bright or dark areas may become compressed and detail lost. Scan lines, dropouts and visible tile mismatches could possibly occur with a failure or calibration problem with the scanning device. Most hardware or software application packages allow the user to adjust brightness and contrast. It should be noted here that an image may appear different from one workstation platform to another. Digital Terrain Model
In aerial photography, all terrain features and objects lean outward from the center of the photograph. This feature is known as relief displacement. During the differential rectification process of the DOI, all objects are mapped to the ground surface or DTM and are therefore corrected for relief displacement. Tall buildings, trees, utility poles, bridges and other features will be projected to the DTM (Figure 3). If these objects are not near the center of the photograph, they will not only lean but may become very distorted after rectification (Figure 1). As mentioned previously, it is very easy to determine if a digital orthophoto meets accuracy requirements. Some problems caused by problems with the DTM usually manifest themselves as shifts, blurring or smearing of the image. Mosaicking
After individual digital orthophotos have been created they should be mosaicked together. Mosaicking is the process in which adjacent images are sutured or seamed together (Figures 2a-b). An overlap area is defined between images and brightness values are compared. A weighted linear function is then used to perform the mosaic. This procedure minimizes sharp differences in contrast and brightness between imagery. If all input imagery were only mosaicked, our end result would be an image database that looked like a checker board. A technique of spatial or histogram equalization is used when some images may be darker or brighter than others. A histogram for gray scale DOI is a graphical representation of the distribution of pixels from 0-255. In this process, a control image is selected and all surrounding imagery is spatially changed to match it. This technique works well as long as there isn't a large variation of shading problems within a photograph. When internal shading anomalies are present, the only course of action is a frequency domain manipulation. An example of a frequency domain manipulation is a low-pass filter utilizing a fast fourier transform (Figures 5a-b). The idea in the low-pass filter is to suppress the frequencies causing the radiometric problems. This process takes knowledge and experience as it could be detrimental if not performed properly. As another example, a high-pass filter could be used for edge enhancement.
Mosaicking and subsequent image processing may remove many of the radiometric problems, but now the geometric problems of mosaicking should be addressed. Each DOI has an associated accuracy. It would seem that joining two images together, a tolerance would now be twice the individual accuracy. As an example, consider a road that runs between adjacent images. The road may be shifted 1 foot in one image, and 1 foot in the opposite direction in the adjacent image. Individually each DOI satisfied accuracy but now a 2 foot difference is present. This edgematching problem is usually small enough that mosaicking smooths the edge match. As another example, imagine that a building is leaning opposite directions in adjacent photographs. A ghosting or collapsing effect may now result on the building in the DOI (Figure 4). After mosaicking, it may become further distorted. Conclusion
The digital orthophoto is becoming an integral part of most GIS implementations today. As can be discerned, the original aerial photography is extremely important to the final DOI product. The image quality issues of DOI are extremely subjective and must be quantified at the outset of each project and mutual expectations derived. The first delivery is typically too late to make a significant change in the product quality without great expense and delay. Before beginning a project, both the purpose of the imagery and the typical uses must be understood. Will features such as fire hydrants, utility poles, manholes or property boundaries need to be identified? Determining potential uses of the data from the user population will determine the functionality and requirements for the imagery. Most users expect imagery that is both radiometrically and geometrically seamless and have very little tolerance for any imagery problems. Quantifying and removing the subjective nature of imagery analysis will allow the project to be a success for all project participants and users. One methodology for establishing project criteria is to set up a scoring method in order to remove the subjective nature of imagery interpretation and acceptance. The model allows the client to determine which categories are most important and weigh them. Each category is then defined in great detail and tolerances established. Each deliverable image, whether softcopy or hardcopy is then scored and, based on the weighing system, passed or failed. If the criteria for acceptance are known prior to delivery, minimal failure rates should be expected. About the Author:
Mark A. Klimiuk is the digital imaging director for Analytical Surveys, Inc. He may be reached at 719-593-0093. Back |
