auto-image-cropper

• Removes extra white boundaries from images to correctly resize canvas.

• Thanks @matthewkmayer for bringing down cropping times (see #1 and #2)!

Screenshots

The borders are just to represent the actual images tested with. (click to zoom)

Installation and Usage

$ cargo install auto-image-cropper

or if you like to live on the bleeding edge

$ git clone https://github.com/Ritiek/auto-image-cropper
$ cd auto-image-cropper
$ cargo build --release

Use ./target/release/autocrop to start using the tool.

USAGE:
    autocrop [OPTIONS] --input <LOCATION>

FLAGS:
    -h, --help       Prints help information
    -V, --version    Prints version information

OPTIONS:
    -i, --input <LOCATION>     Location of input image/directory
    -o, --output <LOCATION>    Location of output image/directory

For example:

$ ./target/release/autocrop -i benchmarking/face.jpg -o face.out.jpg

Python Bindings

This tool also provides Python bindings via PyO3 (using Rust FFI).

You need a Rust nightly toolchain to proceed (PyO3 does not support Rust stable at the moment). If you have rustup, run:

$ rustup default nightly

to switch to nightly channel.

This feature can then be enabled by passing --features "python-binding" to cargo when compiling.

For example:

$ cargo build --release --features "python-binding"

This will generate a dynamic library (*.so) on Linux machines with the name ./target/release/libauto_image_cropper.so.

Let's move this dynamic library into our current working directory:

$ mv target/release/libauto_image_cropper.so .

It can now be utilized via Python scripts using:

>>> import libauto_image_cropper

# Returns the optimal top-left and bottom-right corner
# coordinates for a given image to be cropped
>>> (top_left, bottom_right) = libauto_image_cropper.calculate_corners("benchmarking/face.jpg")
>>> print(top_left)
(442, 73)
>>> print(bottom_right)
(783, 536)

(I haven't checked this out on Windows or OSX, but should follow a similar procedure)

Benchmarks

• This tool was hackishly re-written in Python to compare with Rust - just for fun.

• The benchmarks were done on a MacBook Air running macOS Sierra 10.12.2.

Image	Python	Rust	Times Faster
face.jpg	0.867s	0.155s	5.59
square.png	1.682s	0.142s	11.84
flowers.jpg	2.222s	0.476s	4.66
human.jpg	2.362s	0.294s	8.02
pets.jpg	5.366s	1.704s	3.14
agent47.jpg	51.559s	7.519s	6.85

• Python struggles to find the optimal coordinates but is quick (quicker than Rust) when saving the cropped image back to disk. Rust really outperforms while finding the optimal coordinates.

[2020 / 03] UPDATE: These benchmarks were done in 2017. Rust and its image libraries have a come long way ahead and I believe should now offer even better performance! I'll update these benchmarks when I'm able to.

License

The MIT License