Cruz, Santiago
2 AUTOMATIC INNER ORIENTATION OF NON METRIC 35 AND 70 mm IMAGES
2.1 Inner orientation of non-metric images
As mentioned, the inner orientation of non-metric images will be approached by means of measurements of the frame
corners. These corners can define the coordinate system. But when a frame corner is magnified, both optically (film
based image) or zoomed in (digital image), is difficult to determine the exact position of that corner because the image
is not well defined and sometimes it’s blurred. So, it is usual to make redundant measurements at the four format edges
finding the best regression fit lines for those edges. Once the edge lines have been defined, corner frames are computed
by intersection of lines. Then, the intersection of both diagonals locates the format center (indicated principal point,
IPP). The photo-coordinate system is centered at the IPP and, usually X-axis is forced to be parallel to the lower edge
(figure 1).
u (columns)
rs |
NOn a
€ Measured point at the format edge
O Computed corner
Fit regression line
IPP: indicated principal point
Xy, Yyy photocoordinate system
(parallel to lower format edge)
u, v: image coordinate system (pixel)
(rows)
Figure 1. Coordinate systems in digitized images and explanation of inner orientation process in non-metric images.
Although this methodology has shown to be efficient and it is routinely used, it is very cumbersome and affected of
some uncertainties. Some of the drawbacks can be resumed in a larger number of points to be measured (efficiency is
diminished), final results can be influenced by the number of points per edge, edge irregularities, etc. Moreover, most
commercial software does not allow plotters do the inner orientation by measuring the edges. In order to avoid most of
these problems, an approach using digital image analysis is proposed. Thus, the automatic detection of the frame edges
with high redundant data is possible and lines defining edges and corners are computed in a much more objective way.
Manual and subjective measurements are overcome. Even it is possible drawing marks on the computed position (with
any image analysis software), allowing for inner orientation with a conventional digital plotter.
Photographs have to be previously digitized, well from the film well from paper prints. But, frame edges have to be
clearly visible, so no cropping is allowed. Till recent years this was a real problem, because scanners for slide or
negative films cropped the edges and the total surface frame were not scanned. In any case, Warner, et al. (1992)
employ successfully such scanned images (with cropped edges) and compare results with paper prints. At present, there
are low price desktop scanners equipped with transparency trays, which allow scanning the whole format without any
cropping at high resolutions (up to 1200 dpi). Some of them can even be employed for medium precision
photogrammetric works (Baltsavias and Waegli, 1996).
2.2 Algorithm for automatic frame edge detection
An algorithm for automatic frame edge detection and further inner orientation of photographs has been implemented
under I.D.L.? 5.0 (Interactive Data Language, Research System Inc.). IDL has been chosen because a high capacity for
working with digital images, a powerful and fast computation and a comfortable widget environment. The algorithm is
based in the fact that, usually, an edge is a well defined line if contrast is satisfactory. A directional filter is used for
detecting the edge. We have tried to use the simpler approach, so other complex approaches, such as Hough filters, have
not been considered. Another operators, such as Sobel or Roberts, were also tested but they extract edges in all
directions. Prewitt (directional) filters can extract preferentially horizontal (upper and lower) and vertical (left and right)
lines, so these operators were considered more convenient (figure 2).
When the program starts, this one opens the main window where the image can be loaded (figure 3). In a first step, the
user can extract the rectangular areas where the edges are found, although the program can search for default areas
(spending more time). On these extracted areas would be necessary the application of the four templates (figure 2)
which means designing four algorithms for all edges. With some simple operations only one algorithm needs to be
implemented.
150 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B5. Amsterdam 2000.
Fig
Figure