Context Adaptive Binary Arithmetic Coding Patents: Who Holds the Ground

CABAC is an entropy-coding method used in video compression standards to represent encoded video data more efficiently by adapting probability estimates to the local context of each symbol. In patent filings it appears almost exclusively as part of a video encoder or decoder claim, which is why 91.8% of the 5,413 records in this dataset carry an H04N pictorial-communication classification. Claims typically cover specific ways of deriving, reducing, or updating the contexts and probability states used during binarization, rather than the arithmetic coding engine itself.

Filing activity is concentrated: the five leading assignees together account for 46.1% of the 5,413 records in scope, and the ten leading assignees account for 59.3%. The single leader alone holds 1,493 records, well ahead of the rest of the ranked field. This pattern is typical of standards-essential technology areas, where a handful of large video-codec contributors file heavily across incremental variations to build licensing and cross-licensing positions.

Filings peaked at 896 in 2012, during the H.264/AVC and early HEVC period, and have not returned to that level since. However, the trend has not gone flat: from 2021 to 2024, the last year with a complete publication record, filings rose from 207 to 245, an 18% increase. Figures for 2025 and 2026 look lower only because publication typically lags filing by about 18 months, so those years are still filling in rather than showing an actual decline.

The United States receives the largest share of filings among the offices tracked, at 1,455 records, followed by the European Patent Office at 625 and the WIPO/PCT route at 419. India, China and South Korea follow at 314, 263 and 198 respectively. This spread reflects both major consumer-electronics manufacturing markets and jurisdictions used for international filing strategy via PCT before national entry.

The densest claiming sits inside H04N (video coding, 91.8% of records) and H03M (coding and code conversion, 14.3%), so most defensible new claim space sits outside those cores. Subclasses tied to AI-based computing (G06N, 0.5% of records) and general transmission (H04B, 0.9%) are lightly claimed relative to the field, suggesting model-driven or channel-conditioned approaches to context selection are less contested than classical fixed-context schemes. That said, low density does not guarantee freedom to operate — it means fewer filings to check, not zero.