Convert 4:4:4 to 4:2:0 in VP9 Encoding

This article provides a direct guide on how to accurately convert YUV 4:4:4 video sources to the YUV 4:2:0 chroma subsampling format during a libvpx-vp9 encode using FFmpeg. Standard conversion methods often introduce color bleeding, blurriness in fine details (like text), or color shifts. By utilizing high-quality scaling algorithms and explicitly defining color matrices, you can achieve a highly accurate downsampling process that preserves visual fidelity.

The Challenge with Standard Conversion

When converting from YUV 4:4:4 (full color resolution) to YUV 4:2:0 (quarter color resolution), half of the horizontal and vertical color information is discarded. If you rely on FFmpeg’s default conversion, it typically uses a bilinear scaler, which causes neighboring colors to bleed into each other. To prevent this, you must use a high-quality resampling algorithm and ensure the color space (such as BT.709) is strictly preserved.

The Accurate Conversion Command

To convert your video accurately, use the following FFmpeg command structure:

ffmpeg -i input_444.mp4 -vf "scale=sws_flags=lanczos+accurate_rnd:sws_dither=a_dither,format=yuv420p" -c:v libvpx-vp9 -crf 30 -b:v 0 -colorspace bt709 -color_primaries bt709 -color_trc bt709 output.webm

Parameter Breakdown

To understand why this command ensures an accurate conversion, here is a breakdown of the critical parameters:

• sws_flags=lanczos+accurate_rnd: This is the most crucial part of the conversion. It forces FFmpeg’s software scaler to use the Lanczos algorithm instead of Bilinear. Lanczos preserves sharp edges and prevents color bleeding, while accurate_rnd ensures highly accurate rounding calculations.
• sws_dither=a_dither: Enables automatic dithering. This prevents color banding in gradients when reducing color information.
• format=yuv420p: Explicitly sets the output format to 8-bit YUV 4:2:0. If you are encoding a 10-bit video, change this to format=yuv420p10le to retain the higher bit depth.
• -colorspace bt709 -color_primaries bt709 -color_trc bt709: These flags flag the output file with the correct BT.709 color metadata (standard for HD video). This ensures that web browsers and media players interpret the colors exactly as they were in the source, preventing washed-out colors.
• -c:v libvpx-vp9 -crf 30 -b:v 0: Initiates the VP9 encoder in Constant Quality (CQ) mode, which is the recommended encoding method for libvpx-vp9.

Alternative: Using the zscale Filter

If your FFmpeg build is compiled with --enable-libzimg, you can use the zscale filter, which uses the highly regarded zimg library. It offers even greater precision for color space conversions and downsampling:

ffmpeg -i input_444.mp4 -vf "zscale=matrix=709:filter=lanczos,format=yuv420p" -c:v libvpx-vp9 -crf 30 -b:v 0 -colorspace bt709 -color_primaries bt709 -color_trc bt709 output.webm

The zscale filter performs internal calculations in a higher bit depth, minimizing rounding errors and producing the cleanest possible 4:2:0 output from a 4:4:4 source.