The Residual Arrives as Bits A Block Splitting Claim Written at the Bitstream Layer

This application claims a decoding method in which a block of a video frame is split into thin sub-blocks, each predicted with the same prediction rule, and then corrected using a residual that the decoder recovers by reading information out of a bitstream. The bitstream is the point. Everything else in claim 1 is shared word for word with three companion applications filed on the same day, and this is the one that insists the correction data arrive as coded bits rather than as an abstract block of numbers. It was filed by B1 Institute of Image Technology, Inc. of Korea, with Ki Baek Kim as sole inventor.

The Problem

A video decoder is not handed pixels. It is handed a stream of bits, and the very first thing it does is unpack them. The specification’s own decoder diagram makes this concrete: an incoming image bitstream goes to an entropy decoder, which turns it into coefficients and decoding information that the rest of the machine can use. Every clever idea in a codec has to survive that packing and unpacking, and, more to the point, has to be worth the bits it costs.

That is the constraint the partitioning scheme in this specification is designed around. The document distinguishes two ways of splitting a block. In tree based partitioning, the shape is worked out from several separate pieces of information: whether a split happens, which tree type, which direction, repeated at each level. In index based partitioning, the shape is chosen in one shot by a single index. The specification is candid about the bit economics of the second option. Because the number of bits needed to represent index information grows with the number of supported candidates, the scheme suits single layer partitioning, where the split happens once and the resulting sub-block is not divided again. Deep recursive splitting would make the index too expensive.

The document then spends pages on the coding of that index. It gives worked binarisation tables showing alternative bit patterns: a flat two bits per candidate, or a scheme where the first bit says whether any split happens at all and the remaining bits pick among the split types only. It also handles what happens when a candidate becomes unavailable, because the block sits on a picture boundary or falls below a minimum width or area. The candidate group is rebuilt without the unobtainable type and binarisation is performed again on the smaller group, so a choice that once needed two bits may now need one flag. This is a specification written by people counting bits.

The wider problem is the one the background states directly. Multimedia use has grown with the internet and portable terminals, image processing systems need better performance and efficiency, and existing methods have not kept up. The stated objects are intra prediction performed on a subblock basis, and a way of partitioning into subblocks and choosing the order they are coded in.

What This Invention Does

Claim 1 is an image decoding method with five steps.

A coding block is generated using tree based partitioning, defined as splitting a square block into four equal squares. That block is then split again by index based partitioning, in one of two defined ways. The first partitioning takes a 4N by 4M coding block and produces four identical 4N by M sub-blocks or four identical N by 4M sub-blocks: four flat strips or four narrow columns. The second partitioning splits a block into two equal parts, horizontally or vertically. A prediction block is generated by performing intra prediction on each sub-block, and the claim requires that every sub-block use the same intra prediction mode. A residual block is generated. The coding block is reconstructed from the prediction block and the residual block.

The distinguishing clause follows: the residual block is generated by decoding information on the residual block included in a bitstream. The encoder side, claim 5, mirrors it, stating that the coding block is encoded by encoding information on the residual block into a bitstream. Claim 7 goes one step further and claims a method of transmitting a bitstream, ending with the act of transmission itself. The abstract sets the tone for the whole document by describing the embodiments as encoding and decoding methods, including bitstreams generated from such methods.

The final wherein clause is the one that ties the geometry back to the bit budget. Whether the first partitioning or the second partitioning applies is determined by the size of the coding block. That is a derivation, not a transmission. The decoder already knows the size of the block it is working on, because it followed the tree that produced it, so the choice between four strips and two halves costs nothing to convey. Dependent claim 4 covers the other route, where the partitioning type is determined by an index signalled for the coding block, or alternatively by a property of the block. Dependent claims 2 and 3 add that whether index based partitioning happens at all can depend on a property of the coding block, that property being its area, width or height. Read together, the dependent claims describe a sliding scale between sending information and inferring it, which is the central design tension in any coding tool.

The body supplies the reason the sub-block split earns its keep. When a large block is predicted as a single unit, its bottom right region must be predicted from pixels neighbouring the entire parent block, which are a long way off. Predicted as a sub-block instead, that region can draw on neighbouring sub-blocks already reconstructed in the chosen coding order, so much closer pixels are available. Better prediction means a smaller residual, and a smaller residual means fewer bits, which is the payoff the whole structure is arranged around.

Key Features

• Residual recovered from coded bits. Claim 1 does not treat the residual as a given array of numbers. It requires the decoder to generate it by decoding information on the residual block that is carried inside a bitstream.
• Matching encode and transmit claims. Claim 5 requires the coding block to be encoded by writing residual information into a bitstream, and claim 7 claims the further act of transmitting that bitstream, so the family covers the stream itself and not only the machines at either end.
• One shot index instead of a recursive tree. The sub-block shape is selected by a single index rather than assembled from a sequence of tree decisions, which is why the specification restricts it to a single non recursive partitioning layer.
• Binarisation designed around the candidate group. The specification sets out alternative bit patterns for the partition index, including a scheme where the first bit signals whether a split occurs and the rest choose among split types only.
• Candidate groups that shrink and rebinarise. When a partition type is unobtainable because of a minimum size condition or a picture boundary, the candidate group is reconfigured without it and binarisation is redone, so the removed option stops costing bits.
• A size derived choice that needs no signalling. Which of the two index based partitionings applies is determined by the size of the coding block, information the decoder already holds, so that decision travels for free.

Who Is Behind It

B1 Institute of Image Technology, Inc. is a Korean company whose business is video coding research and the patent portfolio that comes out of it rather than shipped hardware or software. It is named on the licensor list for the HEVC and VVC patent portfolio licence, the pool now administered by Video Codec Licensing as a subsidiary of Access Advance, which is the commercial context in which a company of this kind operates. Ki Baek Kim is the sole named inventor here and appears across the great majority of the company’s filings.

The specification states its lineage in full in the first paragraph of the body, and four Australian numbers are involved. This application is a divisional of Australian application 2024205220, itself a divisional of 2022204573, itself a divisional of 2019247240, which is the Australian national phase of an international application filed on 1 April 2019 claiming the benefit of Korean patent application 10-2018-0037812 filed on 1 April 2018. The international application number is printed as PCT/K2019/003777. The published family record gives it as PCT/KR2019/003777, published as WO 2019/194485, so the Australian document appears to carry a typographical slip in an otherwise complete chain. That correction comes from the published record, not the specification.

Why It Matters

The tool this family describes is a close relative of Intra Sub-Partitions, adopted into Versatile Video Coding, the standard published by the ITU as Recommendation H.266. A 2018 priority date puts it squarely inside the period when that standard’s intra coding tools were being proposed and argued over.

What is distinctive about this particular filing is where it draws the line through the codec. Four applications were lodged on the same day by the same applicant, with identical titles, identical 133 page descriptions and a claim 1 that agrees word for word except for one clause: how the residual block comes into being. One of the four describes it as pure arithmetic, a subtraction at the encoder and an addition at the decoder. One describes it as the output of an inverse transform. One adds inverse quantisation ahead of that. This one places it at the bitstream, and says only that the residual is generated by decoding information carried in the stream.

That choice has consequences a general reader can see without any patent training. A claim written at the arithmetic layer describes a relationship between blocks of pixels, which is elegant but exists only inside a running decoder. A claim written at the bitstream layer describes something with an independent existence: a file, a broadcast, a stream on a wire. It reaches the encoder that produced the bits and, through the transmission claim, the act of sending them. It is narrower than the arithmetic version, because it demands the residual actually be carried as coded information, but narrower in a direction that makes it easier to point at. That trade off is why one description gets split into four applications. The applicant keeps all four rungs and decides later, once examiners and prior art have had their say, which is worth defending.

There is also a reason the bitstream layer is where video coding disputes usually live. Codec standards are defined by what a conforming bitstream contains and how a decoder must interpret it, not by how an encoder chooses to produce it. A claim phrased in terms of decoding information from a stream is phrased in the language the standard itself uses, and for a company whose assets are patents rather than products, that alignment is the point. Those bits are shortened by entropy coding, typically the context adaptive binary arithmetic coding these standards use, which is exactly the stage the binarisation tables are written for.

Related Concepts

• Variable length code – the coding idea behind the specification’s binarisation tables, where common choices get short bit patterns.
• Standard essential patent – the category a bitstream level codec claim is generally written to fall into.
• Bit rate – the quantity every design decision in this specification is trying to reduce.
• Data compression – the general discipline of removing redundancy that prediction and residual coding belong to.
• Patent Cooperation Treaty – the international route by which this Korean family reached the Australian register.

---

AU 2026202165 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.