Why AV1 Performs Better in Dark and Shadowed Scenes

Compressing dark and shadowed video content has historically presented a major challenge for video codecs, frequently resulting in severe banding, blocking artifacts, and excessive bitrate consumption. The AV1 video codec addresses these issues through a combination of perceptual rate-distortion optimization, flexible block partitioning, integrated film grain synthesis, and advanced in-loop filtering. Together, these tools enable AV1 to preserve critical shadow details and eliminate false contours while maintaining significantly lower bitrates than older standards like H.264 and HEVC.

Luminance-Adaptive Quantization (Delta QP)

The human visual system is exceptionally sensitive to subtle intensity variations in low-light environments. Traditional encoders often treat dark blocks similarly to well-lit ones, resulting in coarse quantization that causes harsh color banding (contouring) across smooth gradients. AV1 uses adaptive Delta Quantization (Delta QP), allowing the encoder to modulate the quantizer parameter at the sub-frame and block level. By allocating a finer quantization step to darker areas, AV1 preserves low-amplitude luminance transitions without needlessly inflating the bitrate across the entire frame.

Film Grain Synthesis

Low-light scenes captured by digital camera sensors naturally exhibit high levels of noise. Traditional codecs treat this noise as high-frequency image data, expending a disproportionate number of bits attempting to encode random variations, which often collapses into muddy, fluctuating compression artifacts.

AV1 solves this with Film Grain Synthesis. During encoding, the encoder estimates the statistical parameters of the noise, strips it from the source to compress a clean, low-entropy base image, and transmits the noise characteristics as lightweight metadata. The decoder then deterministically regenerates the grain on playback. This preserves the natural texture and perceived sharpness of dark scenes at a fraction of the traditional bitrate.

Flexible Block Partitioning and Intra Prediction

AV1 introduces superblocks up to 128x128 pixels, paired with a 10-way recursive partitioning tree capable of splitting blocks down to 4x4 pixels, including non-square configurations (such as 1:2, 2:1, 1:4, and 4:1). In shadowed environments:

• Large, uniform shadow regions are captured efficiently by 64x64 or 128x128 superblocks, minimizing signaling overhead.
• Complex transitions between deep shadows and bright highlights are finely divided using flexible non-square splits to prevent edge distortion.

Furthermore, AV1 incorporates specialized smooth intra-prediction modes (Smooth, Smooth Horizontal, Smooth Vertical, and Paeth predictors). These modes extrapolate boundary pixels into flat regions gradually, directly mitigating the step-like artifacts commonly seen in dark gradients.

Advanced In-Loop Filtering

To ensure clean reconstruction of dark scenes, AV1 employs a multi-stage filtering pipeline:

1. Deblocking Filter: Smoothes unnatural boundaries across block edges, heavily mitigating the grid patterns typical of low-bitrate shadows.
2. Constrained Directional Enhancement Filter (CDEF): Applied directly after deblocking, CDEF identifies the direction of primary edges within each 8x8 block and applies a directional filter. This eliminates ringing and blurriness around high-contrast edges in low-light settings.
3. Loop Restoration Filter: Operating as the final in-loop stage, AV1 uses either a separable symmetric Wiener filter or a dual-parameter Self-Guided filter to restore fine details and correct subtle structural errors introduced during the earlier lossy stages.

By combining granular quantization control, synthetic noise modeling, flexible geometry, and edge-preserving reconstruction filters, AV1 maximizes perceptual fidelity in low-luminance scenes while keeping bit consumption strictly controlled.