oavif: Faster target quality image compression
11 hours ago
- #avif
- #image-compression
- #target-quality-encoding
- oavif is a new tool for target quality encoding in image compression, focusing on speed and efficiency.
- Target quality encoding ensures perceptual consistency by using metrics that represent viewer experience.
- oavif improves upon three core components: the metric (fssimu2), the encoder (libaom), and the convergence algorithm.
- fssimu2 is a faster implementation of SSIMULACRA2, reducing memory usage by almost 40%.
- libaom is used for AVIF encoding due to its speed and consistency, with improvements from SVT-AV1-PSY work.
- oavif's convergence algorithm combines binary search, interpolation, and predictive modeling to reduce search times significantly.
- Predictive modeling in oavif reduces average pass count by over 56% compared to traditional methods.
- Error bounds are used to narrow the search space aggressively, improving performance further.
- oavif is written in Zig and supports high bit depth I/O, ICC profiles, and user-configurable encoder settings.
- Future directions include more accurate predictive modeling and potentially keeping a history of buffers for non-convergent cases.