B frame, I frame, and P Frame

Frame B has four reference modes in the MPEG-4. If both the image compression before and after the reference is recorded, the difference between the pixel value of the front image + pixel value of the back image)/2 is recorded, that is, the difference between the image and the average image. Therefore, the number of Difference values recorded is the same as that of P frames. There is only one difference, which is not increased.

Because B
The frame is located in the middle of the front and back frames. It uses the "average of the front and back frames", that is, the "median of the front and back frames", as the predicted value (to predict the pixel of frame B ).
What is the value? If there is an error, record the difference value). In this way, the predicted value is closer to the current real B.
The frame value can be imagined that the difference value to be recorded will be very small or even not required to be recorded at all, so a lot of bits can be saved and the compression rate can be improved.

For example, brightness change-> I B P 7 8 9

If B only references the compression of the previous image, the difference value of 1 must be recorded. If (I + P)/2 is compressed, the difference value is
0. The difference value does not need to be recorded. (Although we need to record two vectors, we can make further predictions and compress the vectors. In general, it will be much smaller than simply referring to the compression of the previous image .)
What should we do? Normally, the screen changes like this. If it is not the case, we will not use frame B. Even if it is not the case, another way is to change the frame B to refer to the image P.
If there are many frames, how can we find them? The probability of using a square with a smaller error can also be found in frame B (because there are many objects to select and reference). Therefore, frame B is better than frame P.
The compression ratio of frames is high. (And the gap is very large)

In addition to compression ratio, the effect of B frame on image quality ..... yes, because B frames are similar to the features of the front and back frames, they have an interpolation effect, so noise can be reduced.

B frame in the MPEG-4, is also very powerful, in addition to the previous three reference modes, as well as Direct Mode, even vector records are saved. Although
4 Mv in the MPEG-4 function, can record four vectors, but the encoder will judge when compression, In the end is the use of 4 Mv
Is the result of the compression small, or is the result of the traditional method small?

If the results of the traditional method are small, the traditional method is used to record the results. If the results of the 4MV method are small, the 4MV method is used to record the results.

(Ps. 4MV will not be used in backward prediction)

You can observe VirtualDub
The blue line shown on the screen during compression, you will find that there is usually a very short blue line between the Blue Line and the blue line inserted in the middle, resulting in a gap, and the gap is very large, this is the B between P
The compression power is more obvious if DivX 5 is used, because DivX 5 can only use IBPBPBPB... one B to one P
The blue lines on the screen are arranged in this way.