Affine Candidate Lists Built From Configured Candidates Alone A Second Claim Set on OPPO's Merge Mode Work

This application is one of a pair of divisionals filed on the same day from the same parent, covering video decoding using an affine candidate list. Where its sibling requires the list to contain a neighbouring candidate as well, this one claims the method with a configured candidate alone. The applicant is Guangdong OPPO Mobile Telecommunications Corp., Ltd.

The Problem

Video codecs compress by describing how blocks of pixels move between frames rather than resending each frame. The motion description itself has to be transmitted, and doing that explicitly for every block is expensive, so codecs build a list of candidate motion values that encoder and decoder can construct identically, and transmit only an index into it.

Standard candidate lists draw on neighbouring blocks, either spatially within the frame or temporally from another frame. That works while nearby blocks are moving the way the current block is moving. It fails when they are not, and it fails particularly on motion that is not a simple slide: a rotating object, a camera zoom, a plane receding into the distance. Affine motion models were introduced to handle exactly those cases, describing a block’s movement using two or three control point vectors rather than one translation vector.

Affine models raise their own candidate problem, though. A neighbouring block may not have used affine mode at all, or may have used a different parameter count, so the pool of directly reusable affine candidates from neighbours can be thin or empty. A codec that can only inherit affine candidates is therefore limited by what happens to sit next door.

What This Invention Does

The claimed method decodes a video signal by constructing an affine candidate list for the current block in which the list comprises a configured candidate. An affine candidate index is parsed from the bitstream, a control point vector for the current block is determined from the list and that index, a motion vector is determined from the control point vector, and affine mode-based prediction is performed using it.

Information about the type of affine mode is parsed from the bitstream, indicating whether a four-parameter or a six-parameter model applies. The configured candidate is determined from a combination of at least two control point vectors: two when the four-parameter mode applies, three when the six-parameter mode applies.

The significant point is what the claim does not require. Its sibling application, AU 2026202057, requires the affine candidate list to comprise a configured candidate and at least one spatial or temporal candidate. This one requires only the configured candidate. That makes it the broader claim of the pair, covering a decoder that constructs affine candidates by combination without necessarily inheriting anything from a neighbouring block.

Key Features

• Configured candidate list. The affine candidate list comprises a configured candidate, without a requirement for spatial or temporal candidates.
• Index parsing. An affine candidate index is parsed from the bitstream to select from the list.
• Control point derivation. The control point vector of the current block is determined from the list and the index.
• Affine mode type signalling. Bitstream information indicates whether four-parameter or six-parameter affine mode applies.
• Parameter-dependent combination. The configured candidate uses two control point vectors in four-parameter mode and three in six-parameter mode.
• Affine mode prediction. Prediction of the current block is performed using the resulting motion vector.

Who Is Behind It

The applicant is Guangdong OPPO Mobile Telecommunications Corp., Ltd., the Dongguan-based consumer electronics manufacturer. The named inventor is Ki Baek Kim. Like its sibling, this application is a divisional of Australian application 2024219401, which descends from application 2019292471 and international application PCT/KR2019/007981 filed in July 2019.

Why It Matters

Filing two divisionals on the same day from the same parent, differing in a single limitation, is a deliberate and well-established practice. Broad claims cover more but are more vulnerable to prior art; narrower claims are more defensible but easier to design around. Prosecuting both in parallel lets an applicant find out which survives examination without betting everything on one formulation, and it is a pattern worth recognising when reading a patent database, because two near-identical entries usually mean a claim-scope strategy rather than a duplicate filing.

The underlying technology is worth the effort because video is the dominant consumer of internet bandwidth, and affine prediction is one of the genuinely notable additions in Versatile Video Coding, the standard finalised in 2020. Handling rotation and zoom natively rather than approximating them with many small translation vectors produces meaningful savings on the kind of camera movement that is common in phone video, which is a direct commercial interest for a handset manufacturer.

Related Concepts

• Affine transformation – the geometric model underlying control point vectors.
• Versatile Video Coding – the standard that adopted affine motion prediction.
• Motion vector – the basic unit of motion description in a codec.
• Divisional patent application – the filing mechanism that produced this pair.
• Data compression – the broader field these techniques belong to.
• IP Australia patent examination – the process both claim sets must pass through.

---

AU 2026202058 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.