246 
The earliest stereo correlators, such as the Stereomat I, developed in 1960 by 
one of the authors at the Hunting Survey Corporation of Toronto Canada, did 
not address the coherence problem.. As a result, such correlators could only 
utilize relatively coarse image detail in rough areas, and would fail com- 
pletely where such detail was not present. 
The second generation of correlators, such as Stereomat II performed first de-- 
gree image transformation of the sample areas which allowed coherence to be 
extended substantially with a corresponding increase in the accuracy and reli- 
ability of the correlation process. The Ares correlator, developed by one of 
the authors at the Itek Corporation of Lexington Massachusetts USA, employed 
second degree transformation to rectify automatically a pair of convergent 
photographs into a viewable stereogram but without measurement facility. 
Polynomial transformation involves several degrees of freedom and correlation 
proceeds stepwise with the correction of each coefficient improving the vis- 
ibility of the other coefficients. 
First and second generation stereo correlators for photogrammetry were essen- 
tially single point correlators, the point being the center of the sample ar- 
ea. They were designed to guide the floating mark of a conventional plotting 
instrument in the production of planimetric and contour manuscripts and strip 
orthophotographs. 
The third generation stereo correlator abandoned image matching by polynomial 
transformation in favor of transformation by direct element displacement. The 
resulting Gestalt Photomapper also departed from the single point concept by 
delivering a full window of dx values at a density of one dx value for every 
64 pixels. 
4 THE GESTALT PROCESS 
The gestalt correlator is so called because the German word describes exactly 
the accumulating memory of form or shape that is essential to its operation. 
The method was invented and the name adopted by one of the authors in 1966 at 
the Itek Corporation. The original purpose of the invention was for the active 
shaping of large parabolic mirrors for use in an orbiting telescope. 
The gestalt process depends upon synchronous raster scanning of the stereo im- 
ages. Useful matching of raster and stereo anisotropies is achieved by making 
the scanning spot vector precisely parallel to the X axis or base line so that 
x disparity between the images becomes a simple time disparity between the 
video signals. 
There are six essential features of the gestalt process: 
1) The two stereo photos are correlated in homologous windows or patches 
about 500 pixels square. 
2) The input video spectrum is divided into frequency bands each having a 
frequency ratio of 2:l. Typically six or seven bands are required. 
3) The iterative matching sequence of the sensing and clearing of local 
disparities starts with the lowest frequency band, representing large image 
areas, and proceeds from band to band until the highest frequency band 
is reached. In the Rastar correlators the highest frequency band 
represents data from individual pixels.. 
4) The resultant dx error signals for each band are combined into a total dx 
error stream which is accumulated synchronously in a serial gestalt memory 
running at frame rate.