4.2.3 Rectification of the vertical aerial photographs

The resultant geo-rectified IKONOS satellite image was then used as a base map from which to attempt the rectification of the 1970s Russian APs. This was first undertaken by Dominic Powlesland, using his excellent GSYS package. Dominic fitted the twenty separate scanned images, using where possible the central portions of the images, to create a photo mosaic, which was then rectified onto the satellite image using a number of control registration points (Fig. 16 - opens GIS). This was done relatively rapidly, and produced a workable image. It corresponded well to the satellite image in places, although noticeably diverged from it in others, and there were clearly problems with mis-matched joins (especially where the coverage was less good).

To explore the problems with this fit, a confidence model was developed by Ceri Rutter as part of an MSc dissertation project at UCL (Rutter 2003). Ceri undertook a new process of transformation:

  1. rectifying each of 15 TIFF images separately using the ArcView imagewarp extension (the other 5 images were of inferior quality and wholly contained within the coverage of the other images)
  2. trimming the images in Idrisi
  3. fitting the images into four rows (reflecting flight paths) in Adobe Photoshop (the tiled image using the best part of the collection)
  4. re-warping the rows in ArcView
  5. pasting together the rows in Idrisi using the 'Concat' mode
  6. exporting to a single TIFF image (Fig. 17 - opens GIS).

Some problems with the image arose, however, from 'boxing' owing to resolution changes in the transformation process (every time an image was saved in Photoshop or warped in ArcView) (Rutter 2003, 37). This is only visible in scales of 1:2,000 or less, but of course is very problematic for the interpretation of the images as it is exactly at such scales that one is attempting to undertake transcription. The problem arose because the image was too large to warp as a single image: the software could not cope with the calculations and crashed when this was attempted (loc. cit.).

In order to assess the accuracy of this new image, 540 control points were identified on both the tiled aerial photograph and the IKONOS image. Each point on the two images was then compared, to measure the distance and direction of difference, and a contour map of these distances produced (Rutter 2003, 40-45). This suggested that the most accurate part of the image was in the northern part of Sultan Kala, where the correspondence between the images was commonly within 1-2m, but could still be up to 5-6m in some areas.

The decision, therefore, in advance of more work on rectification (or perhaps better still, the acquisition of some new vertical images), was that the IKONOS satellite image would be used as the base map for the digitisation process. The positions of all lines on the GIS transcriptions are based upon this image. However, the 1970s Russian vertical APs have considerably more detail, so the process was developed of toggling between the different images, examining each image's potential to help interpret the archaeological evidence, but using the satellite image as the locator for all digitisation.

A series of alternative layers/images were constructed in the GIS to hold different versions of the rectified images (the Powlesland and the Rutter versions).


© Internet Archaeology/Author(s) URL:
Last updated: Mon Sept 29 2008