Core Architecture & Concepts
AV1 builds upon a hybrid block-based coding framework, enhanced with more flexible structures and parallelism. This section introduces the fundamental building blocks of the codec.
AV1 Encoding Pipeline
Input Frame
The process starts with a single video frame.
Partitioning
Frame is divided into Superblocks (up to 128x128), recursively partitioned.
Prediction
Each block is predicted using Intra (same frame) or Inter (other frames).
Transform & Quantize
Residual is transformed to frequency domain, precision reduced.
Entropy Coding
Quantized values compressed into the final bitstream.
Advanced Prediction Techniques
Prediction is key to reducing redundancy. AV1 introduces a vast toolkit of new prediction methods to handle diverse content types, from natural video to screen captures.
Directional Modes
AV1 expands from VP9's 8 directional modes to 56 modes, including finer angles. This allows for much more accurate prediction of edges and textures.
Chroma from Luma (CfL)
Leverages correlation between luma (brightness) and chroma (color) by modeling chroma pixels as a linear function of reconstructed luma pixels.
Intra Block Copy (IBC)
Extremely effective for screen content. Allows a block to be predicted by copying a previously decoded block from within the same frame.
Palette Mode
Designed for content with few colors. Identifies a small palette of unique colors and encodes each pixel by its palette index.
Smooth & Paeth Predictors
Smooth prediction modes for gradients. Paeth predictor excels at corners and sharp edges by selecting the best of three neighboring values.
Recursive Filtering
FILTER_INTRA modes treat blocks as 2D Markov processes, using 7-tap filters on small patches for highly accurate predictions.
Warped & Affine Motion
Models complex movements like zoom, rotation, and shearing on both global and local levels for better dynamic scene predictions.
Compound Prediction
Combines predictions from two reference frames. Includes weighted averages and wedge-partitioned prediction with line pattern blending.
Overlapped Block MC (OBMC)
Reduces blocking artifacts at motion boundaries by extending motion vectors to overlap with neighbors and blending overlapping regions.
Extensive Reference Frames
Can reference up to 7 different frames from a buffer of 8 (past and future), including special-purpose Alternate Reference Frames.
Subpixel Motion
Achieves 1/8-pixel accuracy for luma with adaptive interpolation filters (smooth, regular, sharp) for each direction independently.
Switch Frames
Special inter-frame type enabling seamless resolution switching in adaptive streaming without requiring a full keyframe.
Efficient Coding & Quality Enhancement
After prediction, AV1 uses advanced transform, quantization, and filtering techniques to compress residual data and maximize both efficiency and visual quality.
Transform & Quantization
- Flexible Transforms: Supports both square and rectangular transform shapes (e.g., 64x16, 8x32) up to 64x64, adapting better to object shapes.
- Multiple Kernels: Uses DCT, ADST, Flip-ADST, and Identity transforms to better compact energy for different signal types.
- Quantization Matrices (QMs): Applies different quantization levels to different frequencies based on human perception.
In-Loop Filtering Pipeline
1. Deblocking Filter (DBF)
Smooths sharp artifacts across transform block boundaries.
2. Constrained Directional Enhancement (CDEF)
Removes "mosquito noise" around sharp edges by filtering along identified edge directions.
3. Loop Restoration Filter (LRF)
Uses Wiener and Self-Guided filters to restore detail and reverse compression damage.
Performance & Codec Comparison
The combination of these advanced techniques results in significant compression gains over previous generations. The chart below shows typical bitrate savings for equivalent visual quality.
AV1 Bitrate Savings vs. Predecessors
| Feature | H.264 (AVC) | H.265 (HEVC) | AV1 |
|---|---|---|---|
| Max Block Size | 16x16 | 64x64 | 128x128 |
| Intra Prediction Modes | 9 | 35 | 56+ |
| Motion Compensation | Translational | Advanced | Warped, Affine, OBMC |
| In-Loop Filters | Deblocking | Deblocking, SAO | Deblocking, CDEF, LRF |
| Screen Content Tools | None | Limited | IBC, Palette |
| Royalty Status | Complex Licensing | Complex Licensing | Royalty-Free |