Geometrically Modified Reference Pictures for Global Motion Prediction

The patent applies preliminary action by generating a geometrically modified reference picture before performing inter prediction. The encoder pre-processes the reference picture by applying geometric transformation parameters (such as motion vectors, rotation angles, or scaling factors) to create a modified reference picture that better matches the current picture's geometry. This preliminary geometric modification ensures that when inter prediction is performed, the reference blocks already account for global motion effects, thereby maintaining high prediction efficiency without adding complex processing during the actual prediction step.

The patent implements parameter changes by modifying geometric parameters of the reference picture such as position, orientation, and scale based on motion estimation results. Instead of complex pixel-level warping, the method adjusts key geometric parameters (motion vectors, rotation angles, scaling factors) that define the spatial relationship between reference and current pictures. These parameter changes enable accurate prediction by aligning the reference picture's geometry with the current picture, while keeping the encoding process relatively simple through parameter-based transformation rather than full image reprocessing.

The patent applies feedback by encoding and transmitting geometric modification parameters (motion vectors, transformation matrices, or geometric descriptors) along with the encoded video data. During decoding, the decoder receives these parameters and uses them to reconstruct the geometrically modified reference picture. The feedback mechanism ensures that both encoder and decoder use the same geometric transformation parameters, allowing the decoder to accurately recreate the modified reference picture and achieve high image quality without requiring complex iterative optimization processes.