Artifact Generation: Encoding Errors vs Transcoding Losses

Visual artifacts in digital video trace back to two main causes: encoding errors and transcoding losses. Encoding errors occur when data compression pushes beyond what the codec can accurately represent, leading to distortions like blockiness or color shifts.

On the other hand, transcoding losses appear when already compressed video is reprocessed, compounding small degradations into visible flaws. Understanding how each artifact forms is essential for maintaining consistent visual fidelity across multiple processing stages and preventing avoidable quality loss.

Types of Artifacts

Encoding Errors

Typical artifacts include blocking (grid patterns at low bitrates), ringing (halos near edges), and mosquito noise (flickering distortions in motion areas). These usually appear in high-detail regions, such as textures, during the first encoding pass.

Transcoding Losses

Transcoding introduces cumulative degradation, such as macroblocking becomes more visible, gradients show posterization, and flickering can occur from mismatched motion data. Each transcoding pass increases blurring and reduces sharpness by roughly 10% to 20%.

Causes and Mechanisms

Encoding Errors

These originate from transform coding processes like DCT, where quantization removes fine details to compress data. Limited motion estimation accuracy between frames adds further distortion. At constrained bitrates, complex scenes can lose 5% to 15% PSNR, especially when encoded around CRF 28.

Transcoding Losses

Losses occur when decoding imperfections are reintroduced during re-encoding, including rounding errors and bitrate mismatches. Over successive transcodes, these errors multiply; an initial 2 dB PSNR loss can grow to 4 dB after a second generation, with additional noise accumulating through variable bitrate streams.

Measurement and Detection

Encoding Errors

These are measured using PSNR (typically 30 dB to 40 dB) and SSIM (0.90"0.95). Edge variance and similar metrics help identify blocking artifacts. A VMAF score between 80"90 generally indicates acceptable first-generation quality, with spatial distortion as the main concern.

Transcoding Losses

They are tracked through generational PSNR decay, typically losing 1 dB to 3 dB per re-encode, and SSIM drops of 0.05 to 0.10. Difference imaging between generations reveals cumulative visual loss, and VMAF often drops below 85 after 2 to 3 transcodes, showing temporal and color inconsistencies.

Impact on Quality

Encoding Errors

These reduce visual quality in static or low-light scenes by 10% to 20%, but the effects are limited to the first encoding cycle. In high-motion footage, artifacts can increase by up to 25%, especially in lower bitrate streams below 5 Mbps.

Transcoding Losses

Repeated transcoding leads to progressive quality loss (up to 40% detail reduction after three cycles), making it unsuitable for archival purposes. Dynamic footage is most affected, with artifacts 30% more visible during multi-format conversions and broadcast delivery.

Mitigation Strategies

Encoding Errors

Use higher bitrates (CRF 18"23) or two-pass encoding to minimize quantization loss by around 20%. Enable deblocking filters (-deblock in FFmpeg) and advanced profiles such as High@4.2 to suppress ringing, maintaining SSIM above 0.95.

Transcoding Losses

Reduce transcoding cycles by using direct format conversions and intermediate lossless formats like FFV1. Apply perceptual tuning (-crf 20 and aq-mode 1) to counter cumulative drift. With GPU acceleration, it is possible to limit PSNR loss to under 1 dB per generation.

Comparison Table