AV1 Chroma Subsampling in High-Contrast Text

High-contrast typography often exposes the severe limitations of standard 4:2:0 chroma subsampling, producing noticeable color fringing, haloing, and legibility loss around sharp glyph edges. The AV1 video codec addresses these distortions through specialized architectural enhancements, including Chroma-from-Luma (CfL) prediction, advanced screen content coding tools, flexible block partitioning, and adaptive in-loop filtering. Together, these technologies align color boundaries with sharp luminance transitions, dramatically minimizing visual artifacts around rendered text.

The Challenge of Chroma Subsampling in Typography

In standard digital video, 4:2:0 subsampling reduces the horizontal and vertical resolution of color-difference channels (\(Cb\) and \(Cr\)) by half relative to the luma channel (\(Y\)). While this compression technique is visually acceptable for natural video scenes, it breaks down around high-contrast typography—such as white text on a deep red background or saturated red titles on black. Because color information is mapped at a lower spatial grid, the sharp borders defined in the luma plane lack corresponding color precision, resulting in visible color bleed, jagged contours, and chromatic blur across letterforms.

Chroma-from-Luma (CfL) Prediction

AV1’s primary weapon against subsampling artifacts in text is Chroma-from-Luma (CfL) prediction. Traditional codecs predict chroma blocks strictly using neighboring reconstructed chroma pixels, which causes directional blur across fine lines.

CfL operates on the principle that color boundaries in high-contrast graphics almost always mirror the luminance structure. AV1 reconstructs the full-resolution luma component first, downsamples it to match the chroma grid, removes the DC offset, and then models the chroma values as a linear function of the reconstructed luma pixels. Because the chroma values are directly derived from the sharp edges of the text in the luma channel, color transitions snap tightly to the typographic boundaries instead of bleeding into surrounding pixels.

Screen Content Coding Tools

AV1 integrates specialized Screen Content Coding (SCC) tools engineered specifically for synthetic elements like user interfaces and digital text:

• Palette Mode: For blocks containing text, the gradient-based Discrete Cosine Transform (DCT) often introduces ringing and color smearing. Palette Mode bypasses transform coding entirely, representing the block with a small set of discrete base colors (indexes) for both luma and chroma. This keeps glyph edges razor-sharp and prevents intermediate, muddy chroma values from appearing at the boundary between text and background.
• Intra Block Copy (IBC): High-contrast typography frequently features repeating geometric shapes and recurring alphabet characters. IBC allows the encoder to reference previously decoded text blocks within the same frame as a spatial vector, reproducing crisp character geometry and exact chroma values without recalculating lossy intra-predictions.

Recursive Partitioning and Variable Transforms

Older codecs rely on rigid block sizes, forcing high-contrast text boundaries to share compression parameters with flat background regions. AV1 employs a recursive 10-way partition tree ranging from 128×128 down to 4×4 pixels, including non-square rectangular blocks (such as 1:2, 2:1, 1:4, and 4:1 ratios).

This fine-grained partitioning enables the encoder to isolate narrow typographic strokes into their own transform units. By isolating the text edge, the encoder prevents high-frequency quantization noise from spilling into adjacent flat areas, preserving chroma contrast along the stroke length.

Edge-Preserving In-Loop Filters

AV1 refines edge quality through two sequential filtering stages:

1. Constrained Directional Enhancement Filter (CDEF): Operating directly after the deblocking filter, CDEF identifies the primary directional orientation of edges within each block. It then applies a non-linear filter along the detected angle of the text stroke rather than across it. This smooths out quantization noise and ringing without blurring the high-contrast transition between the text and the background.
2. Loop Restoration: AV1 applies Wiener or self-guided restoration filters across larger frame segments to correct systemic blur introduced during compression. This step sharpens residual softness around characters and restores contrast to fine serifs and thin strokes.

Native Support for Higher Chroma Fidelity

When 4:2:0 subsampling fundamentally cannot preserve the necessary typographic fidelity, AV1 provides native support for 4:4:4 sampling through its Professional Profile. In this profile, the chroma channels retain the identical spatial resolution as the luma channel, eliminating the downsampling step entirely. Combined with 10-bit and 12-bit depth options, AV1 allows graphic-heavy content, screen recordings, and digital interfaces to be compressed with zero subsampling-induced chroma distortion.