When we are taking about prediction we are actually trying to finding relationship or in mathematical terms how related the data are. If there is high amount of correlation, the residue data will be very small. In any compression algorithm it the residue which is encoded and sent across.
Any mathematical operation in video domain must take into consideration the amount of data that is too be proceed. Consider if we take a frame of 144*176 and per pixel is represent by single byte we have 24 Kb per frame. The frame rate for good quality video encoding is 30 frames, so now we have 742 kb of data.
For any processing there is always a meta association which gives the required information for processing. More finely grained operation on image is defined, the higher the information is captured and more the memory requirement. For time being we will keep this factor in mind , which will be use full later .
The following are error estimation methods used in general (taken from wiki)
Squared Error (MSE)
Sum of Absolute Differences (SAD)
Sum of Absolute Transformed Differences (SATD)
Mean Absolute Difference (MAD)
Sum of Squared Errors (SSE)