Recursive splitting of boundary blocks lets image encoders handle non-multiple image sizes while improving compression and neighboring block use.
A five-tap MH-CCP model predicts chroma from collocated and neighboring luma samples to improve video compression while preserving quality.
Regular buffer rules and conformance constraints keep intra block copy vectors valid, simplify processing, and improve screen content coding.
Adaptive binarization for transform skip residuals improves video compression efficiency while controlling bit rate and preserving quality.
Adaptive QP prediction combines neighboring blocks, luma values, and local activity to cut coding cost while preserving video quality.
Adaptive scan orders and transform block sizes improve video decoding efficiency while preserving information needed for accurate coefficient decoding.
A picture-level flag switches chroma cross-decoding on or off, balancing coding efficiency with lower complexity and better parallel processing.
Using separate unrefined and refined motion vectors speeds video decoding while preserving coding gain and reducing blocking artifacts.
Selective DMVD enablement by block balances motion refinement accuracy with decoder complexity in higher-resolution video coding.
Predefined CU-level bi-prediction weights and sign selection improve video compression while reducing weight selection complexity.
Uses neighboring block motion and direction changes to cut motion bit usage while preserving prediction accuracy in video decoding.
A unified MPD structure enables stereo 360 VR zoom switching with SHVC/HEVC streams, reducing complexity and delay under changing bandwidth.
Adaptive boundary CTU depth extends multi-type splitting where frame edges need more flexibility, improving compression and decoding efficiency.
Block-level decoding failure feedback helps exclude corrupted reference areas and prevent image quality loss across subsequent frames.
Shaped spatial and temporal context improves latent-variable coding in hybrid INR compression, cutting bitrate without excessive context complexity.
Motion-based GOP size selection replaces fixed encoding intervals to improve rate distortion performance across changing video scenes.
Dividing video frames into independently coded tiles cuts latency, improves error isolation, and supports parallel encoding and decoding.
A reduced-data reference image stream lets the sender verify that remote images are displayed correctly after long-distance transmission.
Partitioned bidirectional prediction improves image compression accuracy while managing bit precision, weighting, and processing overhead.
Reference-picture-aware motion vector difference signaling improves inter prediction efficiency while reducing coding complexity and bitrate.
Separating motion and residual latents with dedicated synthesis networks improves INR video compression while managing bandwidth and decoding complexity.
Type-based context selection in HEVC decoding cuts memory and processing load by avoiding upper-block dependence where it adds little value.
Secondary transform control data is omitted when valid components use transform skip, cutting redundant signaling and improving image encoding efficiency.
Selected low and high luma-chroma pairs make CCLM line fitting less sensitive to outliers, improving chroma coding efficiency.
Downsampled and upsampled frame sub-block matching improves motion vector accuracy, cutting bit stream size and preserving image quality.
Rule-based video block partitioning disables unsymmetrical trees at size or depth limits to cut coding complexity and improve bitstream efficiency.
Wide-angle intra prediction reuses filtered reference samples across prediction stages to reduce complexity and improve block quality.
Adjusting coded picture buffer size to match encoding level and minimum compression ratio prevents full-frame overflow and underflow.
Partial encoding and error-correction transfer shift multiview coding work to the receiver, cutting transmitter energy use and data load.
Configurable NAL and slice code point mapping merges RAP and non-RAP video streams with lower bitrate cost and reliable decoding.
Constraining temporal motion data fetch positions to the co-located LCU cuts memory bandwidth and cycle overhead in video coding.
Optical flow motion compensation and ROI-based tool selection create consistent datamoshing artifacts without sacrificing coding efficiency.
Parallel affine DMVR processes refine CPMVs and use cost comparison to cut video decoding complexity with minimal coding quality impact.
Selective coded block flag signaling and size-based deblocking improve high-resolution image decoding efficiency and reconstruction.
A derived reference layer flag cuts redundant VPS signaling and speeds reference layer identification in multi-layer video decoding.
Aspect-ratio-based mode mapping extends directional intra-prediction for rectangular blocks, improving coding efficiency with low signaling overhead.
One-bit inclusion flags replace redundant temporal_id signaling, cutting reference picture list bits and avoiding extra list modification signaling.
BP SEI scope control limits sublayer conformance checks in multi-layer bitstreams, reducing coding errors and processor, memory, and network use.
Adaptive CCSO filter unit sizes improve video reconstruction accuracy by balancing regional filtering flexibility with manageable decoding complexity.
Refined subblock motion vectors and linear regression improve affine inter prediction accuracy, boosting coding efficiency and reducing distortion.
Boundary downsampling and selective upsampling improve matrix intra prediction coding efficiency while controlling video processing complexity.
Boundary-aware masked blending and selective MV refinement improve compound prediction when reference blocks extend beyond picture edges.
Predicting future view counts from submission features lets transcoding queues prioritize likely popular videos and reduce delays for high-quality content.
Fixed filters followed by adaptive coefficient filtering reduce residual VVC artifacts and noise while preserving manageable decoding complexity.
Offline-trained machine learning prediction models improve block prediction accuracy and video compression without adding real-time training overhead.
Layer-specific DPB limits let video decoders size reference picture lists by layer type, improving coding efficiency and reducing memory waste.
When frame sizes change, disabling temporal motion vector prediction without same-size references cuts redundant processing and improves coding efficiency.
Selective 4-sample edge filtering on an 8×8 grid cuts HEVC deblocking complexity and memory use while preserving image quality.
Buffered frame-boundary pixels let loop restoration filters improve compression ratio and video quality in video coding.
Intra-coded sub-units are spread across other pictures to enable random access and refresh with lower bit-rate spikes and joining delay.