mini-jpeg

A minimal JPEG decoder implementation written in Rust for educational purposes. The goal is to make the code readable and serve as a learning resource for understanding JPEG compression algorithms, rather than focusing on performance optimization.

Input: Baseline DCT JPEG images in JFIF format (.jpg/.jpeg)
Output: Decoded bitmap images (.bmp)

What This Decoder Implements

The decoder implements the full JPEG decoding pipeline:

JFIF File Format Parsing: SOI, APP0, COM, DQT, SOF0 (Baseline DCT), DHT, SOS, and EOI markers

Decoding Pipeline:

• Huffman Decoding - Constructs Huffman trees and decodes DC/AC coefficients with run-length encoding (RLE)
• Inverse Quantization - Dequantizes DCT coefficients using quantization tables
• Inverse Discrete Cosine Transform (IDCT) - Converts frequency domain back to spatial domain
• Color Space Conversion - YCbCr to RGB transformation
• Image Reconstruction - Assembles 8×8 MCU blocks into the final image

Current Limitations

This implementation supports:

• Baseline DCT JPEG images only (no progressive JPEG)
• 4:4:4 chroma subsampling (no subsampling)
• 3-component YCbCr color images and 1-component grayscale images

Optimization Opportunities

As this is a learning-focused implementation, there are many low-hanging fruit optimizations that could significantly improve performance:

IDCT Optimizations

• Integer-based IDCT: Replace floating-point calculations with fixed-point integer arithmetic
• Separable 1D transforms: Use two 1D IDCT passes (rows then columns) instead of a 2D transform, reducing complexity from O(n⁴) to O(n³)
• Precomputed cosine tables: Cache cosine calculations instead of computing them for each coefficient
• SIMD vectorization: Leverage CPU vector instructions for parallel computation

Parallelization

• RST marker support: Implement restart marker (RST) handling to enable parallel decoding of image segments
• Multi-threaded block processing: Process multiple MCU blocks concurrently

Other Improvements

• Chroma subsampling support: Implement 4:2:2 and 4:2:0 subsampling with upsampling
• Progressive JPEG: Add support for progressive DCT encoding
• Memory optimization: Reduce memory allocations and improve cache locality

Learning Resources

This implementation was developed with the help of the following excellent resources:

• Let's write a simple JPEG library - Koushtav Chakrabarty

• Let's Build a JPEG Decoder - Video Series

• JPEG - Wikipedia

• JPEG File Interchange Format - Wikipedia

Building and Running

# Build the project
cargo build --release

# Run the decoder (default image: images/lenna.jpg)
cargo run --release

The decoded output will be saved as a BMP file in the same directory as the input JPEG.

Project Structure

src/
├── main.rs           # Entry point
├── jpeg_parser.rs    # Main decoder logic and JFIF parsing
├── huffman_tree.rs   # Huffman tree construction and decoding
├── bit_reader.rs     # Bit-level reading utilities
└── constants.rs      # JFIF markers and zigzag ordering

Dependencies

• bmp - For saving decoded images
• byteorder - For reading big-endian data from JPEG files
• anyhow - For error handling