Lycon

A minimal and fast image library for Python and C++.

Lycon is a small subset of optimized image operations derived from OpenCV.

Current set of features include:

• Reading and writing JPEG and PNG images
• Fast SIMD optimized image resizing
• Zero-copy interop with NumPy whenever possible

Tested on:

• Linux (Ubuntu 14.04) with Python2.7.6 and 3.5.2.
• macOS (Sierra, 10.12) with Python 2.7.11 and 3.5.1.

Install

pip install lycon

Native extension dependencies:

• CMake 2.8 or newer
• C++ toolchain
• LibJPEG
• LibPNG

Ubuntu

Single-line command for installing all dependencies:

sudo apt-get install cmake build-essential libjpeg-dev libpng-dev

Anaconda

When working within an Anaconda Python distribution, it is recommended to use the latest cmake version (3.6 or newer). Older versions can lead to a mismatch between the libpng and libjpeg headers used to build Lycon (usually the system headers), and the linked library (which may be preempted by the Anaconda-scoped version). To install the latest cmake version:

conda install cmake

Example

import lycon

# Load an image as a numpy array
img = lycon.load('mittens.jpg')
# Resize the image using bicubic interpolation
resized = lycon.resize(img, width=256, height=512, interpolation=lycon.Interpolation.CUBIC)
# Crop the image (like any regular numpy array)
cropped = resized[:100, :200]
# Save the image
lycon.save('cropped-mittens.png', cropped)

Limitations

Compared to other image processing libraries (OpenCV, pillow, scikit-image), Lycon offers a very limited set of operations. Intended usages include data loaders for deep learning, mass image resizing, etc.

Advantages over OpenCV

• Drastically smaller (at the cost of drastically fewer features)
• Python module installable via pip
• Images use the more common RGB ordering (vs OpenCV's BGR)

However, if you already have OpenCV installed, Lycon's advantages are minimal.

Advantages over PIL(low)

• Faster
• First-class NumPy support
• Full support for floating point images

Advantages over Scikit-Image

• Drastically faster

Benchmarks

• The table below lists execution time (in seconds), averaged across 10 runs
• The multiplier next to the time is the relative slowdown compared to Lycon

• Blank cells indicate that the operation is not supported by the library
• All operations performed on a 16k (15360 x 8640) RGB image
• Tests performed on Ubuntu 14.04 running on an Intel Core i7 (Skylake)
• OpenCV 3.2+ (master: a85b4b5), Pillow 4.0.0, skimage 0.12.3, Python 2.7.3
• OpenCV can potentially achieve better performance with GPU implementations and proprietary libraries like Intel IPP

License

• All code derived from the OpenCV project is licensed under the 3-clause BSD License.
• All Lycon-specific modifications are licensed under the MIT license.

See LICENSE for further details.