Direct mode coding for bi-predictive pictures in the JVT standard

The direct mode can considerably improve coding efficiency of Bi-predictive pictures within the JVT standard since, by exploiting the correlation that may exist between pictures, it can effectively represent block motion without having to transmit any motion information. Until recently the only correlation that was exploited was temporal correlation which, unfortunately, implies that the motion and timestamp information for previous pictures are available in both the encoder and decoder. In this document we introduce an alternative method for calculating the direct mode parameters using spatial correlation. Our method can achieve similar coding efficiency to the temporal method, while at the same time reducing memory requirements and without requiring timestamp information. In addition, improvements on the existing Rate Distortion Optimization related to Bi-predictive pictures within the JVT coded are also presented.