Merge Mode-Based Inter-Prediction Building Better Motion Candidates by Combining the Ones You Already Have

This application covers a video encoding and decoding method that improves how a codec predicts motion, by adding a combined candidate, derived by merging existing candidates, to the list a decoder chooses from. The applicant is Guangdong OPPO Mobile Telecommunications Corp., Ltd.

The Problem

Demand for high definition and ultra high definition video has grown across almost every application, and each step up in resolution multiplies the amount of data involved. Video compression handles that by exploiting the fact that consecutive frames are mostly similar. Rather than sending each frame from scratch, a codec sends instructions describing how a block of pixels has moved since a previous frame. This is inter-prediction, and its efficiency is a large part of what determines the final file size.

The instructions themselves cost bits. Sending an explicit motion vector for every block is expensive, so merge mode was introduced: rather than transmitting the motion information, the encoder builds a list of candidate motion values that the decoder can build identically from data it already has, and sends only an index into that list. A short index replaces a full vector.

That works only if a good candidate is actually in the list. Conventional lists are populated from spatial candidates, taken from neighbouring blocks in the same frame, and temporal candidates, taken from a co-located block in another frame. When none of those neighbours happens to carry motion close to what the current block is really doing, the encoder is forced to fall back on explicitly coding the motion, and the bit saving disappears. Complex motion such as rotation, zoom or shear is exactly where simple neighbour copying breaks down.

What This Invention Does

The method constructs a merge candidate list for the current block, derives the block’s motion information from that list, and performs inter-prediction using the derived motion information. What distinguishes it is the composition of the list: alongside spatial and temporal merge candidates, it includes a combined merge candidate, derived by combining n candidates already in the list.

The specification pairs that with affine motion prediction, which describes a block’s movement using control point vectors rather than a single translation, allowing the codec to represent rotation and scaling rather than only sliding. Information parsed from the bitstream indicates whether a four-parameter or a six-parameter affine mode applies, and the configured candidate is assembled accordingly: from a combination of two control point vectors in the four-parameter case, or three in the six-parameter case.

In this application, the claimed candidate list comprises the configured candidate together with at least one spatial or temporal candidate. Deriving new candidates by combination rather than only inheriting them from neighbours means the list can cover motion no single neighbour exhibits, which is where the stated advantage, improved accuracy of motion information, comes from.

Key Features

• Merge candidate list construction. A candidate list is built for the current block, from which motion information is derived.
• Combined merge candidate. A candidate derived by combining n candidates already in the list is included alongside conventional ones.
• Spatial and temporal candidates. The claimed list also comprises at least one spatial or temporal merge candidate.
• Affine mode signalling. Information parsed from the bitstream indicates whether a four-parameter or six-parameter affine mode applies.
• Control point vector combination. The configured candidate is built from two control point vectors in four-parameter mode or three in six-parameter mode.
• Improved motion accuracy. The stated benefit is better accuracy of motion information than spatial and temporal candidates alone provide.

Who Is Behind It

The applicant is Guangdong OPPO Mobile Telecommunications Corp., Ltd., the Chinese consumer electronics manufacturer best known for smartphones, which has built a substantial video coding patent portfolio. The named inventor is Ki Baek Kim. The application is a divisional of Australian application 2024219401, itself a divisional of application 2019292471, which entered the national phase from international application PCT/KR2019/007981 filed in July 2019.

Why It Matters

Video is the majority of internet traffic, so compression efficiency translates directly into bandwidth, storage and energy at very large scale. Each new codec generation, from H.264 through HEVC to Versatile Video Coding, roughly halves the bitrate needed for the same quality, and it does so through an accumulation of individually modest improvements to prediction, transform and entropy coding.

Merge mode refinements are a good example of that accumulation. No single change to how a candidate list is populated is dramatic, but candidate list design has been one of the more actively worked areas across successive standards precisely because it sits on the critical path of every inter-predicted block. Techniques adopted into a standard also become commercially significant, since every conforming encoder and decoder must implement them, which is why handset manufacturers such as OPPO file and continue to prosecute in this area years after the original priority date.

Related Concepts

• Inter frame – the prediction technique this method improves.
• Motion compensation – the underlying idea of describing a frame in terms of movement.
• Affine transformation – the geometric model behind control point vectors.
• High Efficiency Video Coding – the standard that introduced merge mode.
• Versatile Video Coding – the generation that adopted affine prediction.
• Video codec – the wider system this technique sits inside.

---

AU 2026202057 was published in the Australian Official Journal of Patents on 9 April 2026 and is open for public inspection. Patent applications represent inventions that are sought to be protected and do not necessarily reflect commercially available products.