5× faster fast_blur in image-rs
Summary
This article discusses optimizing the image blur functionality in the image-rs Rust crate, achieving up to 5.9x speedups for u8 images by replacing float-heavy paths with integer accumulators and a reciprocal-based division. It explains blur algorithms (Gaussian, box, and fast_blur), profiling results, and a design using BlurAccumulator to support multiple pixel types, including a merged PR in image-rs. The piece emphasizes performance trade-offs and real-world applicability in image processing workloads.