STATISTICS OF TELEVISION SIGNALS 



757 



SHIFT, AS, NUMBER OF VERTICAL PICTURE ELEMENTS 

 ■100 -80 -60 -40 -20 20 40 60 80 100 



-i — r 



— I — I — I — I — I — I — \ — I — I — I I I I I T" 



SHIFT, AS, NUMBER OF HORIZONTAL PICTURE ELEMENTS 

 -60 -50 -40 -30 -20 -10 10 20 30 40 50 60 



1 1 \ 1 1 1 1 1 1 I I I T 



0.6 



A (AS) 



0.2 j^ 



-0.5 



-0.2 -0.1 0.1 0.2 



SHIFT, As, IN INCHES 



Fig. 5 — Plots of autocorrelation in horizontal and vertical directions for two 

 pictures. 



Fig. 4 shows some pictures for which autocorrelation measurements 

 have been made. The results can be presented in the various ways 

 shown in Figs. 5, 6, and 7. Fig. 5 shows conventional plots of A versus 

 As in the horizontal and vertical directions. Scene B is seen to have more 

 correlation than Scene D, and curve shapes range from remarkably- 

 linear to somewhat like exponential. Fig. 6, gi\ang contours of constant 

 autocorrelation, brings out the variation with the angle 6. Scene A 

 happens to have its greatest correlation in the vertical direction, but 

 that was not found to be a general rule by any means; Scene B, for ex- 

 ample, has its greatest correlation in the horizontal direction. No pre- 

 ferred directions appear to exist in general. In Fig. 7 attention is focused 

 on the more local correlation, for small values of As. The average cor- 

 relation among horizontally adjoining picture elements, designated by 

 Aio , is seen to be approximately 0.99 for Scene B and only 0.75 for 

 Scene C. A20 denotes the correlation for a horizontal spacing of two 

 picture elements while Aoi denotes the correlation among vertically 

 adjoining picture elements. 



It should l)e pointed out that the pictures which gave the above 

 results were not band-limited to the standard 4-mc resolution. However, 

 before the results were used quantitatively, the proper band limitation 

 was applied mathematically. This has the effect of rounding off the peaks 

 of the curves, decreasing the autocorrelation drop within the first Ny- 

 quist interval by up to approximately 24 per cent. 



