Motion compensation method and device using bidirectional optical flow
AssigneeSK Telecom Co., Ltd.
InventorsJeong-yeon LIM, Hyo Song KIM, Hyeong-duck KIM, Se-hoon SON, Jae-seob SHIN, Gyeong-taek LEE, Sun-young LEE
Life of this patent
Estimating the term…
Abstract
Disclosed herein is a method for adaptive bidirectional optical flow estimation for inter-screen prediction compensation during video encoding. The method aims to reduce complexity and/or cost of bidirectional optical flow (BIO) at a pixel level or a subblock level.
Claims9
A video encoding apparatus for predicting a target block using a bi-directional optical flow, the apparatus comprising one or more processor configured to:
- determine the target block, by partitioning a pre-determined size of block in a tree structure;
- generate a first motion vector for a first reference picture and a second motion vector for a second reference picture;
- derive a variable from sample value differences between samples in the first reference picture which are determined based on the first motion vector and samples in the second reference picture which are determined based on the second motion vector, wherein the variable is SAD (Sum of Absolute Differences) or SSE (Sum of Squared Errors) and is used for determining whether to apply the bi-directional optical flow;
- generate a prediction block of the target block from the first and second reference pictures, by selectively either applying or skipping the bi-directional optical flow depending on the variable;
- and
- encode partition information related to the tree structure, and residual signals that are differences between samples in the target block and samples in the prediction block.
A video decoding method for predicting a target block using a bi-directional optical flow, the method comprising:
- decoding partition information from a bitstream, and determining the target block by partitioning a pre-determined size of block in a tree structure;
- reconstructing residual signals corresponding to the target block from the bitstream;
- generating a first motion vector for a first reference picture and a second motion vector for a second reference picture;
- deriving a variable from sample value differences between samples in the first reference picture which are determined based on the first motion vector and samples in the second reference picture which are determined based on the second motion vector, wherein the variable is SAD (Sum of Absolute Differences) or SSE (Sum of Squared Errors) and is used for determining whether to apply the bi-directional optical flow;
- generating a prediction block of the target block from the first and second reference pictures, by selectively either applying or skipping the bi-directional optical flow depending on the variable;
- and
- reconstructing the target block by using samples in the prediction block and the residual signals.
A non-transitory recording medium storing a bitstream generated by a video encoding method of predicting a target block using a bi-directional optical flow, the method comprising:
- determining the target block, by partitioning a pre-determined size of block in a tree structure;
- generating a first motion vector for a first reference picture and a second motion vector for a second reference picture;
- deriving a variable from sample value differences between samples in the first reference picture which are determined based on the first motion vector and samples in the second reference picture which are determined based on the second motion vector, wherein the variable is SAD (Sum of Absolute Differences) or SSE (Sum of Squared Errors) and is used for determining whether to apply the bi-directional optical flow;
- generating a prediction block of the target block from the first and second reference pictures, by selectively either performing or skipping the bi-directional optical flow depending on the variable;
- and
- encoding partition information related to the tree structure, and residual signals that are differences between samples in the target block and samples in the prediction block.
Dependencies are inferred from the claim text.