What Is AV1 Chroma from Luma Prediction?

Chroma from Luma (CfL) is an intra-prediction tool in the AV1 video codec that predicts color detail directly from reconstructed brightness information within the same block. By exploiting the strong correlation between color and brightness variations in natural images, CfL significantly reduces cross-channel redundancy. This mechanism improves compression efficiency, preserves sharp color boundaries, and decreases the bitrate needed to represent chroma data.

In digital video, visual data is typically represented using the YCbCr color space, separating brightness (luma, Y) from color (chroma, Cb and Cr). In traditional codecs, intra-frame prediction processes luma and chroma components largely independently, using neighboring reconstructed pixels to predict the current block's contents. However, physical objects usually exhibit color changes and texture shifts at the exact same spatial boundaries where brightness changes occur. Treating these channels separately leaves substantial correlation unexploited.

CfL addresses this inefficiency by modeling the chroma values as a linear function of the reconstructed luma values inside the same block. Because consumer video formats commonly use YUV 4:2:0 subsampling—meaning chroma has half the horizontal and vertical resolution of luma—the prediction process follows a specific mathematical pipeline:

  1. Luma Downsampling: The reconstructed luma pixels corresponding to the chroma block are downsampled to match the chroma resolution. AV1 supports multiple subsampling methods depending on the chroma subsampling format.
  2. AC/DC Separation: The average value (DC component) of the downsampled luma block is computed and subtracted from each downsampled luma pixel. This isolates the high-frequency variations (AC details) of the brightness texture.
  3. Linear Scaling: The encoder determines a scaling factor (\(\alpha\)) that best correlates the isolated luma variations with the actual chroma variations. This parameter is quantized and transmitted in the bitstream.
  4. Chroma Reconstruction: The final chroma prediction is generated by multiplying the normalized AC luma values by \(\alpha\) and adding the predicted chroma DC value:

\[\text{Chroma}_{\text{pred}} = \text{DC}_{\text{chroma}} + \alpha \times (\text{Luma}_{\text{downsampled}} - \text{DC}_{\text{luma}})\]

By using a simple linear model rather than complex multi-parameter regressions, CfL maintains a low computational footprint during decoding. The decoder only needs to perform downsampling, subtraction, multiplication by \(\alpha\), and an addition.

The primary benefit of CfL is enhanced compression efficiency. It yields noticeable BD-rate reductions—particularly in the chroma planes, where bit savings can reach 5% to 10% on intra frames—while effectively eliminating color bleeding artifacts around sharp edges.