LINKAK PHKDKTIOX IN TKLKVISION 7<S3 



changes in brightness. In (he case of "previous vahie" prediction a sudden 

 change in brightness produces only one error, wiiei'e for "slope," 1\vo 

 errors result. This, to some extent, accounts for tlie lesser amount of 

 powei' reduction measured for these scenes. 



I'ig. 17 shows the appearance of the error signal foi' "previous line" 

 l)re(liction in the three scenes. Where vertical contour lines are pre- 

 dominately left after "previous value" i)rediction shown in Fig. 15, 

 horizontal contours, are more prevalent now. It can be noted that the 

 power reduction for "prc\ious line" prediction is less than that for "pre- 

 \'ious value" prediction. This is due principally to the increased distance 

 of the previous line sample from >S'o,o. If the closest horizontal sample 

 was taken at the same distance from the present value of the signal as 

 the pre^ious line sample, then the power reduction using these signal 

 \'alues individually for prediction would be essentially the same for most 

 pictures. 



Fig. 18 shows the error signal appearance for "planar" pi-ediction. 

 Here, vertical as well as horizontal contours are deleted. In Scene A 

 the ti'ee trunk has almost completely vanished. In Scene B the picture 

 has an extremely flat appearance. Scene C exhibits the lack of hori- 

 zontal and \'ertical contours best, since only sloping contours are left. 

 The power reduction figures at the lower left hand corner also show values 

 foi' minimiun error power. For most pictures, the error power can be 

 reduced by a factor of one-half again over the planar coefficients by 

 modifying the weighting coefficients. The coefficients for this modified 

 planar case are given by 



Sp = 3*^1,0 + 3*^n,l ~ 3*J1,1 



These coefficients generally produce an error signal with less power than 

 the coefficients used for "planar" prediction. 



While all pictures contain redundancy, the error signals from these 

 simple hnear predictors shown in Figs. 15, 16, 17 and 18 can \'isually 

 l)e noted still to contain large amounts of redundancy. The contours of 

 the models and of the various objects are readily identifiable. Were all 

 redundancy removed, the picture would be completely chaotic and would 

 appear very much like random noise, although greater efficiency in 

 transmission would be achieved. For lichei' rewards, more sophisticated 

 methods of prediction will be reciuired. 



ACKNOWLEDGMENT 



The author wishes to acknowledge with grateful ajipreciation the in- 

 valuable guidance of Dr. B. 'SI. ()li\(>r. It was j)rinci])ally through iiisef- 

 orts that this study was made j)ossible. 



