Libwce: A Minimal Entropy Coding Library for Wavelet-Based Image Compression

A developer has released libwce, a bare-bones implementation of entropy coding designed to isolate and clarify a key component of modern image compression. Rather than bundling the entropy layer with complete codec implementations like JPEG, JPEG 2000, JPEG XS, and WebP, libwce focuses exclusively on how wavelet coefficients are converted into compressed bits. The library is implemented as a single Rust file containing approximately 500 lines of code with no external dependencies beyond the standard library.

The library implements a patent-clean Bit-Plane Count (BPC) style entropy layer inspired by JPEG XS. BPC coding works by processing wavelet coefficients—the mathematical representation of image data after transformation from the spatial to frequency domain—in groups of four. For each group, the encoder determines the minimum number of bits required to represent all coefficients, then compresses these bit-plane count values by estimating them from neighboring groups and encoding only the small residual differences. This approach exploits the natural clustering of similar coefficient magnitudes in neighboring image regions.

Libwce employs multiple compression strategies for the bit-plane count values, including RUNNING (delta encoding against the previous group) and ZERO (residual encoding against a lossy bits threshold) predictors. These can be optionally combined with a sparse-block flag for regions containing mostly zero coefficients. The encoder performs a single-pass cost search across multiple Rice coding parameters to select the optimal predictor combination for each frequency band, allowing different strategies to be applied where they work best—such as in textured regions versus flat areas.

The library is designed as a stateless implementation that operates purely on caller-owned buffers, with no hidden state or I/O operations. A complete decoder for a single sub-band requires only three lines of code, demonstrating the simplicity of the API. This minimalist approach allows developers to understand entropy coding in wavelet-based image compression without navigating the complexity of full codec implementations, which typically layer multiple components including transform operations, metadata handling, and standards-specific framing code.