| Digital Ortho
How to Design a Quality Digital Orthophoto
Addressing issues and procedures that need to be understood
to produce a consistent and quality digital orthophoto product.
By Gary Manzer and Kathryn Tiffin Technology today in photogrammetric systems has greatly facilitated the efficient production of high quality digital orthophotos. This in turn is contributing to an expanding market demand for geographic information worldwide. Specialists in geographic information systems (GIS) are recognizing the need to incorporate digital orthophotos as base maps on which to construct the foundation of their GIS or to integrate orthophotos with existing spatial databases. As an example, orthophotos can provide the accuracy required in creating a GIS cadastre database upon which records of land ownership, property tax revenues, utilities, resources and other tiers of information can be constructed.
Aerial photographs and satellite images are generated in a variety of data formats and must undergo numerous mathematical manipulations including corrections, transformations and rectifications (adjustments) to become fully processed and useful georeferenced geographic information. This is the science of photogrammetry and geomatics engineering. This article presents guidelines for digital orthophoto standards which reflect the practical experience of Triathlon Mapping Corp. in over 14 years of creating film based and digital orthophotos for the North American and international market. Here we address some of the issues and procedures that need to be understood to produce a consistent and quality digital orthophoto product.
The two main issues relating to digital orthophotography are image quality and accuracy. Image Quality
These include transparent original materials, the scanner, the rectification and radiometric algorithms selected, and the experience and care given by production personnel.
The issues relating to the transparent originals include:
1) the camera system quality,
2) the sharpness of the transparency, and
3) the density range of the transparency.
The issues relating to the scanner include:
1) its radiometric accuracy,
2) its radiometric range or dynamic range,
3) the sampling rate (scanner resolution), and
4) whether its pixel value output is proportional to the density or to the transmissivity of the media being scanned
5) its Modulation Transfer Function (MTF) resulting in "true" or interpretable image resolution. Positional Accuracy
The positional accuracy of a digital orthophoto is affected by:
1) the geometric accuracy of the scanner,
2) the selection of an appropriate pixel size relative to final scale,
3) the procedure for controlling the photo,
4) the image orientation procedure,
5) the magnification ratio (photo scale divided by final scale),
6) the aerial camera system and lens focal length,
7) the DTM quality used for rectification,
8) the resampling algorithms used,
9) the display and measuring software used on the final ortho imagery. Image Accuracy
A digital orthophoto can be assessed for accuracy only on the horizontal x,y axis unlike modern three-dimensional vector maps. It is created from a scanned aerial transparency by three-dimensional resect and rectification using a Digital Terrain Model (DTM). Any vertical or horizontal errors in the DTM will necessarily appear as planimetric (horizontal) errors in the digital orthophoto. Orientation procedures (both inner and absolute), of the scanned image are also important to the quality of the final product.
The elevation inaccuracies contained in such a model used for orthophoto rectification will be transferred to the resulting digital orthophoto in a predictable way.
For image planimetry containing ground based points or vectors, the resulting digital orthophoto should have an accuracy closely tied to the vector accuracy. Since aerial triangulated ground points have the highest degree of accuracy normally available, they should normally be incorporated into any DTM used for rectifying images. There is always much more image planimetry requiring rectification than there is corresponding DTM vector planimetry. Therefore, the accuracy of the digital orthophoto for discrete points measured in areas which are approximated by the TIN facets of the model, will reflect the inherent errors of these approximated elevation values.
A process called differential rectification removes most of the vertical relief and tilt displacement in the image, however, it is not always possible to remove all vertical relief. This is because on a photograph, areas at above ground elevations lie closer to the camera at the time of exposure and therefore appear larger than corresponding areas lying at lower elevations. The tops of objects are displaced from their bases. This causes any object standing above the ground to lean away from the principal point of a photograph radially. So relief displacement is directly proportional to the height of the object imaged, inversely proportional to the flying height above the ground and directly proportional to the radial distance to the object from the principal point.
The radial distance from the center of the air film scanned of the discrete point being measured on the orthophoto, together with the focal length of the camera lens, can be used to calculate the error that was introduced to the orthophoto planimetry. In other words, the closer the orthophoto planimetry was to the center of the original aerial photo frame the less sensitive it is to DTM error. As orthophoto planimetry gets closer to what was the edge of the original aerial frame, the more the accuracy of the rectification is dependent on the accuracy within the DTM. There will necessarily be orthophoto planimetric errors up to 80 percent of the vertical errors (for 6" photography) in the DTM, for areas at the edge of the double model area. (Orthophotos made in double model format are orthos made from every other 60 percent frame of standard aerial photography.) DEM related positional errors in digital orthophotos can be described more concisely by the general formula:
e(o) = e(DEM) x tan A
where e(o) = the positional error in the orthophoto
e(DEM) = the vertical error in the DEM
A = the viewing angle in degrees outward from the center of the photograph (determined by the distance where ground objects are radially displaced from the photo center, divided by the focal length of the camera).
Therefore, when measuring the planimetric accuracy of an orthophoto, the best result will be achieved if the clearly defined ground object you are measuring is a ground based aerial triangulation point. Other accuracy measurements of discrete ground elevation points which fall on top of DTM vectors will be the next most accurate, and the least accurate discrete points measured within a digital orthophoto will be those areas of the orthophoto which use only the approximation of the TIN model as a base for rectification. The worst result accuracy would be apparent, for discrete image points that were originally imaged at the edge of the original aerial photo, and in the center of a TIN facet where the errors in elevation are the greatest. Scanning Acceptable Materials
Scanning should be carried out only from first or second generation transparent originals in positive or negative form. Scanning from paper prints cannot be allowed due to the reduced resolution of such materials. If original roll materials are to be scanned then the scanner must not physically come in contact with the film while it is in motion. This will help eliminate scratches to the image that could be introduced by the scanner. Scanner Geometric Accuracy
Any scanner to be used for the scanning of aerial imagery must be calibrated using a standard glass plate. A minimum of 20 well distributed points should be read during the geometric accuracy test. It is not acceptable to post process the scan data to achieve this requirement. The calibration report should show the manufacturer's name, model number, serial number, date of calibration, individual readings of grid points, rectangularity, calculated RMSE value for the instrument, and the technicians name and signature. The report must show that the scanner maintains a geometric accuracy of not more than 5 microns RMSE at the scanners plate scale. Scanner Radiometry
Scanning should be carried out either from an autododged second generation transparency or an original negative processed by autododging software. If the former method is used the second generation autododged B&W transparency ideally should be processed to have a density range between 0.3 and 1.3. If the latter method is used (ie. from original negatives), the dodging must be done in the scanner in at least a 10 bit space (for each band if color) prior to outputting the final 8 bit files for storage. The conversion from the original image to digital pixels should be carried out proportional to the density of the original. Representation of the tones proportional to the transmissivity of the original will produce substandard scans.
Images of "normal" scenes (ie. scenes without predominant snow, sand, water, or shadow areas), should display a Gaussian distribution of pixel values with the mean centered on pixel value 127. There should be no unusual spikes of data, nor gaps in individual pixel bins. The minimum and maximum densities should be captured without clipping and without leaving unused bins on the ends of the histogram. Scanners should be capable of scanning original imagery with a density range (Dmax -Dmin) of at least 1.8 without resulting in missing pixel bins or gaps, and without post processing to merge bins. When a standard photographic nine step wedge is scanned the resulting tones should give an RSME ±10 pixel units from their calculated position based on the steps density. Resolution of the Scanner
"True" resolution, defined as what you can resolve from what you see, should not be confused with the scan sample rate set on a scanner (the nominal resolution of the scan). Test scans taken of the same transparency, but from competing scanners, can show vastly different resolution when compared to each other. Disparity in "true" resolution can be so great, that one 15 micron scan can have the same "true" resolution as another scanner's 30 micron scan. This disparity has to do with, among other things, the Modulation Transfer Function (MTF) of the scanner. The MTF is critical in achieving optimal or "true" resolution, but it is a more difficult thing to empirically measure, and therefore must be subjectively judged. In addition, there is also the question of noise which we have not addressed here at all. Choosing Appropriate Scan Sample Rates (nominal resolution)
In all cases the original scan resolution should be at least as fine as the desired pixel size less 20 percent. Scanning at an even finer resolution is also acceptable. This requirement is made in order to ensure that a pixel is not created at a finer resolution than the original scan due to varying photo scales in the original photography. This guideline is designed to allow for the production of high quality hard copies at the final scale. It also means that the image can be comfortably viewed at up to twice the planned scale in soft copy format.
Appropriate scan resolutions can be calculated using the concept of magnification. Magnification is calculated by dividing the original nominal photo scale by the final scale. Magnification ratios for black and white photography should be kept to less than 8; magnification ratios for color photo should be kept to less than 7. This constraint is recommended because the average resolution of most aerial film will not support larger ratios without significant noise being introduced into the image. The following examples outline the methodology. (See Tables 1 and 2) Image Orientations
The camera calibration report for the camera in use should be input to the system prior to orientations commencing. This should include fiducial distances, radial distortions, and calibrated focal length. In addition, the parameters for Earth curvature correction and refraction should be applied. Choosing Appropriate Final Pixel Sizes
If the digital ortho is to be created accurate to a final scale, the maximum size of the pixel necessary to identify ground based objects in order to determine whether the stated accuracy has been met, can be calculated. Rectification Algorithms
Image rectification should be carried out using either cubic convolution, an equivalent, or better algorithm. Nearest neighbor or bi-linear convolution are not acceptable. This is due to the artifacts that are introduced into the resulting image by the nearest neighbor and bi-linear convolution methods. Mosaicking
Where it is not possible to complete a sheet from a single image, multiple images will have to be mosaicked together. Prior to completing a mosaic join, the required images should have their tones balanced, preferably by an autododging method that does differential tone and contrast adjustment throughout the images. The join line between the images in use should be defined as a polyline. This line should be chosen so as to minimize the obtrusiveness of the join itself. If feathering is used along the join line it should not result in any noticeable image degradation such as image blurring or double imagery. Orthophoto Accuracy
All orthophotos should meet the following minimum accuracy standards. Ninety percent of all well defined features on unobscured ground should fit within 0.35 mm at the final map scale. Images that tie to each other should appear continuous in geometry within ± 0.7 mm at the final map scale. If the control and DTM were provided by the client and these accuracy conditions can not be met, this should be reported to the client prior to completion of the project. Compression
Ideally all digital orthophotos should be delivered in uncompressed format to allow the greatest flexibility to the end user. Most image compression algorithms are lossee, that is to say when they are compressed they can never be decompressed to get the exact same image without loss of image data. J-peg compression of more than a four or five to one compression ratio for most "average" B&W scenes will start to cause noticeable loss of information. As the rate of compression is dependent upon the characteristics of the image, the compression ratio will vary throughout the project. The compression factor however should remain consistent throughout a project. Quality Control
All completed images should have the corrected positions of the ground control (pug points or "burned in" image crosses) verified against the calculated position from the AT process. A report should be made on any points outside of the allowable tolerances for orthophoto accuracy defined below. Adjacent images should be checked to ensure imagery is continuous without gaps. Linear features within an image should be checked for fit to existing map details if available. At the join of sheets, features should be checked to ensure that they tie within the allowed tolerance as well.
In addition, a visual inspection should be done of the image. Areas of concern are imagery of inconsistent tone relative its surroundings, areas of apparently smeared or blurred imagery. Image smears are often caused by a spike in the DTM which should be fixed. Smears can also be caused by extreme relief relative the angle of view from the given photo. Large areas of such image smearing may be cause for rejection. In such a case an intermediated photograph, more closely centered on the smeared area should be rectified if available and mosaicked into the final sheet. If no other image is available for repair of the smeared area then the end user should be informed. A contractor's only other option, after consulting with the client, is to smooth the DTM in the affected area so as to reduce the extent or severity of the smearing. This will result in reduced planimetric accuracy in this area and is thus is to be kept to an absolute minimum so as not to unnecessarily degrade the image accuracy.
Photogrammetrists are by education and experience, best equipped to determine the range of accuracies contained within a digital orthophoto project, and to advise their clients as to the costs and benefits associated with any particular methodology. It is important to note that there must be a fairly rigorous relationship between the final pixel size and the stated final scale of an orthophoto. The user must be able to clearly identify objects at final scale to even assess whether they are within the stated accuracy specification. Such a rigorous relationship has been outlined here. About the Authors:
Gary Manzer is marketing manager and Kathryn Tiffin is geodesy and geomatics engineer for Triathlon Mapping Corp. in Burnaby, B.C., Canada. They may be reached at 604-294-8861 (phone) or 604-294-6521 (fax). |
