xbrz-rs

Crates.ioxbrz-rs
lib.rsxbrz-rs
version0.1.0
sourcesrc
created_at2024-07-28 16:15:09.114113
updated_at2024-07-28 16:15:09.114113
descriptionA high quality image upscaling algorithm designed to preserve key details in low-resolution pixel art.
homepage
repositoryhttps://github.com/bell345/xbrz-rs
max_upload_size
id1318055
size90,082
Thomas Bell (bell345)

documentation

README

xbrz-rs

Crates.io version docs.rs GPL-3.0 license

This project is a Rust port of the C++ implementation of the xBRZ pixel scaling algorithm, originally created by Zenju. You can download the original C++ version on SourceForge. Both the C++ version and this port are licensed under the GNU General Public License v3.

See the example images section for a comparison of this library's output with nearest neighbour scaling.

Usage

There is currently one public function called scale_rgba, which can scale an RGBA format image from 2x up to 6x the original resolution:

use std::path::Path;

fn upscale_image(input_path: &Path, output_path: &Path, factor: usize) -> Result<(), image::ImageError> {
    let in_image = image::open(input_path)?;

    let width = in_image.width();
    let height = in_image.height();

    let rgba = image::RgbaImage::from(in_image);
    let out_rgba = xbrz::scale_rgba(
        &rgba,
        width as usize,
        height as usize,
        factor
    );

    let out_width = width * factor as u32;
    let out_height = height * factor as u32;

    image::save_buffer(
        output_path,
        &out_rgba,
        out_width,
        out_height,
        image::ExtendedColorType::Rgba8,
    )
}

Features

To compare pixels, the RGB values are converted into YCbCr representation. To do this efficiently, this package creates a lookup table (LUT) which converts between an RGB difference and scalar difference computed with YCbCr coordinates.

By default, this lookup table uses compressed 5-bit per channel indices such that the LUT only uses 128 KiB of memory. However, by enabling the large_lut feature, the full range of 8-bit differences is used, which causes the LUT to take up 64 MiB of memory. The difference will be negligible, but you may enable it if you wish for higher accuracy.

Example images

Also see the test/images directory for examples at more scaling factors.

Sample I

Nearest Neighbour x3 xBRZ algorithm (xbrz-rs) x3
Sample one, nearest neighbour scaled 3x Sample one, xBRZ scaled 3x

Sample II

Nearest Neighbour x3 xBRZ algorithm (xbrz-rs) x3
Sample two, nearest neighbour scaled 3x Sample two, xBRZ scaled 3x

Yoshi

Nearest Neighbour x6 xBRZ algorithm (xbrz-rs) x6
Yoshi, nearest neighbour scaled 6x Yoshi, xBRZ scaled 6x
Commit count: 0

cargo fmt