AV1 Largest Transform Block Size Explained

This article explores the transform block architecture of the AV1 video codec, focusing specifically on the maximum allowable transform size defined in the standard. It details the dimensions of the largest transform block, how it integrates with AV1's superblock structure, and the performance optimizations implemented to maintain encoding and decoding efficiency at higher resolutions.

In the AV1 specification, the largest transform block size allowed is 64x64 samples. While AV1 supports coding block partitioning (superblocks) up to 128x128 samples, the primary transform operations are capped at 64x64 to balance compression efficiency with computational complexity.

Supported Transform Dimensions

AV1 supports both square and rectangular transform shapes. The 64-point dimension is utilized in several transform configurations:

• Square transforms: 64x64
• Rectangular 1:2 transforms: 64x32 and 32x64
• Rectangular 1:4 transforms: 64x16 and 16x64

These extended sizes allow the encoder to effectively capture low-frequency energy in flat or smooth areas, which is particularly beneficial for 4K and 8K video content.

High-Frequency Truncation

To manage hardware complexity and memory bandwidth, the AV1 standard enforces coefficient truncation on 64-point transforms. When a 64x64 transform is applied, only the lowest 32x32 frequency coefficients are retained and coded; the remaining high-frequency coefficients are automatically zeroed out.

Similarly, for rectangular blocks with a 64-sample dimension (such as 64x32), the 64-point side retains only the first 32 low-frequency values. This truncation significantly reduces the operations required during inverse transform processing without noticeably degrading visual quality, as large blocks are predominantly chosen in areas lacking high-frequency detail.