Non-zero rounding and prediction mode selection techniques in video encoding

摘要

In one aspect of this disclosure, rounding adjustments to bi-directional predictive data may be purposely eliminated to provide predictive data that lacks any rounding bias. In this case, rounded and unrounded predictive data may both be considered in a rate-distortion analysis to identify the best data for prediction of a given video block. In another aspect of this disclosure, techniques are described for selecting among default weighted prediction, implicit weighted prediction, and explicit weighted prediction. In this context, techniques are also described for adding offset to prediction data, e.g., using the format of explicit weighted prediction to allow for offsets to predictive data that is otherwise determined by implicit or default weighted prediction.

主权项

1. A method of decoding video data, the method comprising:
receiving an encoded bitstream; entropy decoding the encoded bitstream to generate quantized residual coefficients; inverse transforming the quantized residual coefficients to generate a residual block; determining a first prediction block form a reference frame in a first list; determining a second prediction block from a reference frame in a second list; generating a weighted prediction data that depends on at least the two prediction blocks from the reference frames in the two or more lists of data and includes at least two offsets and a rounding adjustment, wherein the weighted prediction data is generated approximately according to the following equation:
pred(i,j)=(pred0(i,j)*w0+pred1(i,j)*w1+((o1+o2)<>(r+1), wherein pred(i,j) is the weighted prediction data, pred0(i,j) and pred1(i,j) are the first and second prediction blocks, w0 and w1 are weight factors, o1 and o2 are the at least two offsets, and r and 2r are the rounding adjustment and provide the rounding in conjunction with a right shift operation “>>” by (r+1) bits; adding the residual block and the weighted prediction data to obtain a reconstructed block; and outputting the reconstructed block as decoded video data.