RealVideo 4

• FourCC: rv40
• Company: Real
• High level whitepaper: https://datatype.helixcommunity.org/2003/codecdocs/rv9_tech_overview_v1_5.pdf
• Samples: http://samples.mplayerhq.hu/real/VC-RV40/

RealVideo 4 is available in RealPlayer starting with version 9. RealVideo 10 (introduced with RealPlayer 10) uses the same RV40 video format, and can be played by RealPlayer 9 decoder. All Realvideo 10 enhancements are in the encoder.

RV40 is somewhat of a prototype of H.264. It also shares the same internal structure as RealVideo 3, the main difference only in coding methods and codes used.

Technical Details

The following block diagram is lifted wholesale from the above whitepaper and gives a general overview of the general decoding flow for a RealVideo 4 data stream.

This actually is not very interesting as it describes many video codecs.

Transforms

These are the 4x4 inverse transforms used in the RV40 decoding process. These transforms assume an input coefficient matrix laid out as:

 c0  c1  c2  c3
 c4  c5  c6  c7
 c8  c9 c10 c11
c12 c13 c14 c15

ITransform4x4

For each row [c0 c1 c2 c3]:

c0' = 13*c0 + 17*c1 + 13*c2 +  7*c3
c1' = 13*c0 +  7*c1 - 13*c2 - 17*c3
c2' = 13*c0 -  7*c1 - 13*c2 + 17*c3
c3' = 13*c0 - 17*c1 + 13*c2 -  7*c3

For each column [c0 c4 c8 c12]:

c0' = (13*c0 + 17*c4 + 13*c8 +    c12 + 512) / 1024
c4' = (13*c0 +  7*c4 + 13*c8 - 17*c12 + 512) / 1024
c8' = (13*c0 -  7*c4 + 13*c8 + 17*c12 + 512) / 1024
c12'= (13*c0 - 17*c4 + 13*c8 +    c12 + 512) / 1024

Note that if only the DC coefficient is non-zero, it is possible to simply compute n as:

n = (DC * 169 + 512) / 1024

and then populate the transformed matrix with n in all 16 positions.

Intra16x16ITransform4x4

For each row [c0 c1 c2 c3]:

c0' = 13*c0 + 17*c1 + 13*c2 +  7*c3
c1' = 13*c0 +  7*c1 - 13*c2 - 17*c3
c2' = 13*c0 -  7*c1 - 13*c2 + 17*c3
c3' = 13*c0 - 17*c1 + 13*c2 -  7*c3

For each column [c0 c4 c8 c12]:

c0' = (624*c0 + 816*c4 + 624*c8 + 336*c12) / 32768
c4' = (624*c0 + 336*c4 - 624*c8 - 816*c12) / 32768
c8' = (624*c0 - 336*c4 - 624*c8 + 816*c12) / 32768
c12'= (624*c0 - 816*c4 + 624*c8 - 336*c12) / 32768

Prediction

See H.264 Prediction for descriptions of RV40's prediction modes which are very similar to those found in H.264.