Template matching for video coding

摘要

In one aspect of this disclosure, template matching motion prediction is applied to B-frames. In another aspect of this disclosure, template matching motion prediction as applied to video block coding may include generating a template offset, generating a weighted sum of absolute differences, selecting a number of hypotheses used to encode video blocks based on the cost associated with the number of hypotheses and signaling, with a new syntax, to a decoder, the number of hypotheses used in encoding, rejecting hypotheses if the difference in value between a hypothesis and a reference hypothesis is greater than a threshold value, and/or generating the content of a sub-block that does not have reconstructed data available by combining motion-compensated prediction and luma residuals.

主权项

1. A video coding method for coding a current video block of a current video unit, the method comprising:
generating a first set of hypotheses for template-matching predictive video coding, wherein the hypotheses in the first set are based on individual frames of data from at least one of a list containing previous frames in display order and a list containing future frames in display order; generating a second set of hypotheses for template-matching predictive video coding, wherein the hypotheses in the second set are based on bi-directional motion prediction between one or more pairs of frames wherein each pair of frames comprises one frame from the list containing previous frames in display order and one frame from the list containing future frames in display order; wherein each of the hypotheses in the first and second sets of hypotheses comprises video data corresponding to a template shape defined relative to a video block location; applying techniques for selecting one or more hypotheses from among the first and second sets of hypotheses, wherein the techniques for selecting are based at least in part on a number of hypotheses to use for coding the current video block, and wherein the number of hypothesis is identified from a predetermined set of numbers of hypotheses; coding, via a video coder, the current video block using predictive video data from the selected hypotheses to produce a coded video block; and generating a syntax element that indicates the identified number of hypotheses, wherein the syntax element is provided to a video decoder to decode the coded video block.