Crate convolve2d[−][src]
Expand description
This crate defines an easy and extensible way to conduct image convolutions, in a way that
is free of system dependencies, and works with no_std
.
The purpose of convolve2d
is to provide a single package that provides everything you need to
conduct image convolutions suitable for computer vision or image manipulation. Here’s a breif
overview of what’s on offer:
-
Two convolution functions: allowing you to pass your own buffer if speed is important, or have a buffer allocated and returned for a more idiomatic interface.
-
Traits: Convolution is defined generically across the
Matrix
trait. If you have a custom image type, simply define an implementation ofMatrix
, and you’re good to go! -
Built-in
image
Support: We also offer support for theimage
library through a feature flag (disabled by default), allowing you to seamlessly use the types you’re already used to! -
rayon
: Compute convolutions in parallel using therayon
flag. (Enabled by default) -
no_std
Operation: to suit the needs of specialty systems or WASM. -
Kernel Generators: The
kernel
module provides generation functions for a number of kernels commonly used in image processing.
While other convolution libraries may be more efficient, using a faster algorithm, or running on the GPU, this library’s main focus is providing a complete convolution experience that is portable and easy to use.
Features:
The following features are supported:
Feature | Default | Description |
---|---|---|
std | Yes | Allow access to the standard library, enabling the DynamicMatrix type. |
rayon | Yes | Use rayon to compute convolutions in parallel. |
image | No | Add extensions for interoperation with the image crate. |
full | No | All features. |
To use the library in no_std
mode, simply disable all features:
convolve2d = { version = "0.1.0", default-features = false }
Notes on image
Compatibility
Compatibility with the image
library is provided using the image
feature flag. This flag
provides the following features:
-
The various pixel formats (
Rgb
,Luma
, etc…) can now be converted to and from theSubPixels
type. This allows them to be scaled and added as required for convolutions. -
ImageBuffer
can be converted to and fromDynamicMatrix
es usinginto
andfrom
. -
ImageBuffer
s for which the pixel type isLuma
can be used asMatrix
es directly. This is because each element in the underlying data structure is one pixel. (Whereas in an RGB image, each element is one subpixel, meaning we need to group withSubPixels
)
Modules
Definitions for various kernels that can be generated automatically.
Structs
A concrete implementation of Matrix
for which the size is not known at compile time.
A Matrix
with a size known at compile time.
A collection of subpixels that should make working with multi-channeled images more convenient.
Traits
An easily implementable interface for types that can be used in a convolution.
Functions
Perform a 2D convolution on the specified image with the provided kernel.
Write the convolution of the provided image and kernel into the specified buffer.