NanoJPEG++

NanoJPEG++ is a project created with the goal of optimizing the original NanoJPEG library to achieve performance close to that of libjpeg-turbo. This version of the library is developed in C++ and focuses on delivering high-speed JPEG image decoding. NanoJPEG++ is a header-only project, making integration into your project straightforward.

Description

NanoJPEG++ is a modern, high-performance library for JPEG image decoding. This version includes several enhancements to achieve a significant performance boost compared to the original NanoJPEG library:

• Bitstream Processing: Reading data from the stream in 32-bit chunks allows simultaneous extraction of Huffman codes and their symbols. This speeds up decoding by reducing the number of shift operations.
• Optimized Lookup Table: The Huffman table has been reduced and optimized to use less memory. It is split into DC and AC components with different bit depths. If a direct lookup fails, the number of leading ones is computed, and a quick transition is made to the nearest Huffman code to optimize the search.
• IDCT Vectorization: Modern compilers have achieved automatic vectorization of the IDCT (Discrete Cosine Transform) process, eliminating the need for complex vector operation code or assembly inserts.

NanoJPEG++ is a header-only project, which means you do not need to build separate binaries or libraries. Simply include the header files in your project to take advantage of the library's features.

Changes from NanoJPEG

In the original NanoJPEG library, RGB conversion was a significant performance bottleneck. In NanoJPEG++, I chose to omit the RGB conversion and focus solely on decoding JPEG components. For color space conversion, you will need to use other specialized libraries.

Example Usage

Here's how you can use NanoJPEG++ to decode a JPEG image:

#include "nanojpeg.h"

int main() {
    const uint8_t* buf; // JPEG data buffer
    size_t size;        // Buffer size

    // Assume buf and size are already initialized
    auto result = nanojpeg::decode(buf, size);

    // Access the decoding results
    int width = result.width;
    int height = result.height;
    auto& components = result.components;
    bool is_ycck = result.is_ycck;
}

Performance

NanoJPEG++ (with clang-15 -Ofast -flto) shows an average performance similar to libjpeg-turbo (with SIMD assembly). The performance of the library depends on the specific hardware and compiler used. However, it is designed to be highly efficient and suitable for embedded systems and real-time applications.

• The app.cpp file in the project contains usage examples and can be used for performance testing.

License

NanoJPEG++ is licensed under the MIT License. See LICENSE for more details.

Contributing

We welcome contributions to the project. If you would like to make improvements, please create a pull request or report issues in the GitHub Issues section.

Contact

If you have any questions or suggestions, you can reach us at vovach777@gmail.com.