cykooz.resizer 2.0

A fast image resizer

cykooz.resizer

cykooz.resizer is package with optimized version of image resizing based on Rust's crate fast_image_resize.

CHANGELOG

Installation

python3 -m pip install cykooz.resizer

Or with automatically installing Pillow:

python3 -m pip install cykooz.resizer[pillow]

Information

Supported pixel types and available optimisations:

• U8 - one byte per pixel:
  • native Rust-code without forced SIMD
  • SSE4.1 (partial)
  • AVX2
• U8x3 - three u8 components per pixel (e.g. RGB):
  • native Rust-code without forced SIMD
  • SSE4.1 (partial)
  • AVX2
• U8x4 - four u8 components per pixel (RGBA, RGBx, CMYK and other):
  • native Rust-code without forced SIMD
  • SSE4.1
  • AVX2
• U16x3 - three u16 components per pixel (e.g. RGB):
  • native Rust-code without forced SIMD
  • SSE4.1
  • AVX2
• I32 - one signed integer (32 bits) per pixel:
  • native Rust-code without forced SIMD
• F32 - one float (32 bits) per pixel:
  • native Rust-code without forced SIMD

Implemented resize algorithms:

• Nearest - is nearest-neighbor interpolation, replacing every pixel with the nearest pixel in the output; for upscaling this means multiple pixels of the same color will be present.
• Convolution with different filters:
  • box
  • bilinear
  • catmull_rom
  • mitchell
  • lanczos3
• Super sampling - resizing an image in two steps. First step uses the "nearest" algorithm. Second step uses "convolution" with configurable filter.

Usage Examples

Resize Pillow's image

from PIL import Image

from cykooz.resizer import FilterType, ResizeAlg, Resizer


resizer = Resizer(ResizeAlg.convolution(FilterType.lanczos3))
dst_size = (255, 170)
dst_image = Image.new('RGBA', dst_size)

for i in range(1, 10):
    image = Image.open('nasa_%d-4928x3279.png' % i)
    resizer.resize_pil(image, dst_image)
    dst_image.save('nasa_%d-255x170.png' % i)

Resize raw image with alpha channel

from cykooz.resizer import AlphaMulDiv, FilterType, ImageData, PixelType, ResizeAlg, Resizer

def resize_raw(width: int, height: int, pixels: bytes):
    src_image = ImageData(
        width,
        height,
        PixelType.U8x4,
        pixels,
    )
    alpha_mul_div = AlphaMulDiv()
    resizer = Resizer(ResizeAlg.convolution(FilterType.lanczos3))
    dst_image = ImageData(255, 170, PixelType.U8x4)
    alpha_mul_div.multiply_alpha_inplace(src_image)
    resizer.resize(src_image, dst_image)
    alpha_mul_div.divide_alpha_inplace(dst_image)    
    return dst_image

Change used CPU-extensions

from cykooz.resizer import FilterType, ResizeAlg, Resizer, CpuExtensions


resizer = Resizer(ResizeAlg.convolution(FilterType.lanczos3))
resizer.cpu_extensions = CpuExtensions.sse4_1
...

Benchmarks

Environment:

• CPU: Intel(R) Core(TM) i7-6700K CPU @ 4.00GHz
• RAM: DDR4 3000 MHz
• Ubuntu 20.04 (linux 5.13)
• Python 3.9
• Rust 1.59.0
• cykooz.resizer = "2.0"

Other Python libraries used to compare of resizing speed:

• Pillow = "9.0.1" (https://pypi.org/project/Pillow/)

Resize algorithms:

• Nearest
• Convolution with Bilinear filter
• Convolution with Lanczos3 filter

Resize RGBA image 4928x3279 => 852x567

• Source image nasa-4928x3279.png

Package (time in ms)	nearest	bilinear	lanczos3
Pillow	0.92	112.37	273.12
cykooz.resizer	0.52	64.97	119.94
cykooz.resizer - sse4_1	0.52	23.20	36.92
cykooz.resizer - avx2	0.52	17.27	26.35

Resize grayscale (U8) image 4928x3279 => 852x567

• Source image nasa-4928x3279.png has converted into grayscale image with one byte per pixel.

Package (time in ms)	nearest	bilinear	lanczos3
Pillow	0.29	38.30	117.40
cykooz.resizer	0.21	20.36	36.66