API reference

OpenCV.jl wraps the OpenCV computer-vision library for Julia.

Images and matrices are represented by Mat, a 3-dimensional AbstractArray that shares memory with OpenCV's cv::Mat. Mind the axis order: a cv::Mat is exposed with axes (channels, cols, rows) — see Mat for details and the consequences for descriptor/result matrices.

The bindings are generated from the OpenCV headers (see gen/); the full list of wrapped methods is in gen/funclist.csv.

Mat{T,M,D} <: AbstractArray{T,3}

The image/array type of OpenCV.jl: a 3-dimensional AbstractArray view that shares memory with an OpenCV cv::Mat (no copy).

Memory layout

A cv::Mat is exposed to Julia with axes (channels, cols, rows) — that is, size(m) == (nchannels, width, height). This is OpenCV's row-major height × width × channels buffer reinterpreted in Julia's column-major order, so the bytes are shared with C++ directly.

Things to remember:

• A grayscale H×W image has size == (1, W, H).
• A 3-channel H×W image has size == (3, W, H).
• Result matrices follow the same rule. For example, an ORB descriptor matrix of N keypoints × descriptorSize bytes has size == (1, descriptorSize, N) — the keypoint axis is dimension 3, not 1.

Construction

Mat(data) wraps an existing AbstractArray{T,3} without copying; most OpenCV functions return Mats converted from C++.

Examples

julia> using OpenCV

julia> data = reshape(collect(UInt8, 1:24), 1, 4, 6);   # (channels, cols, rows)

julia> m = OpenCV.Mat(data);

julia> size(m)
(1, 4, 6)

Vec{T,N,C,D} <: AbstractArray{T,1}

A fixed-length, 1-dimensional AbstractArray view backing OpenCV's small fixed vectors (cv::Vec{T,N}) — for example the per-pixel channel tuple of a multi-channel Mat. Vec{T,N}(data) wraps a length-N array (throwing DimensionMismatch otherwise); Vec{T,N}(cxxobj) wraps a C++ object's storage in place.

Point{T}(x, y)

A 2-D point with coordinates x, y (OpenCV cv::Point_). Pass the element type explicitly to match the wrapped signature, e.g. Point{Int32}(10, 20) for pixel coordinates or Point{Float32} for sub-pixel ones.

Point3{T}(x, y, z)

A 3-D point with coordinates x, y, z (OpenCV cv::Point3_).

Size{T}(width, height)

A width × height size (OpenCV cv::Size_). Note this is width-then-height, unlike a Mat's (channels, cols, rows) axes. Many functions expect Size{Int32}, e.g. Size{Int32}(640, 480).

Rect{T}(x, y, width, height)

An axis-aligned rectangle with top-left corner (x, y) and the given width and height (OpenCV cv::Rect_).

RotatedRect(center::Point{Float32}, size::Size{Float32}, angle::Float32)

A rectangle rotated by angle degrees about center (OpenCV cv::RotatedRect).

Range(start, end_)

A half-open integer index range [start, end_) (OpenCV cv::Range). end_ is spelled with a trailing underscore because end is a Julia keyword.

TermCriteria(type, maxCount, epsilon)

Iteration stopping criterion for iterative algorithms (OpenCV cv::TermCriteria): stop after maxCount iterations and/or once the change falls below epsilon, according to type (a combination of TermCriteria_COUNT / TermCriteria_EPS).

cvComplex{T}(re, im)

A complex number with real part re and imaginary part im (OpenCV cv::Complex).

load(f::File, [flags])
load(s::Stream, [flags])

FileIO integration: decode an image file or stream into an OpenCV Mat. Registered for BMP/JP2/JPEG/PNG/TIFF, so FileIO.load("img.png") returns a Mat with the usual (channels, cols, rows) layout. The optional flags argument is an imread-style colour flag (e.g. OpenCV.IMREAD_GRAYSCALE).

save(f::File, image, [params])
save(s::Stream, image, [params])

FileIO integration: encode an OpenCV image (a Mat or any InputArray) to a file or stream. The format is taken from the File/Stream type; the optional params::Vector{Int32} is forwarded to imwrite/imencode (e.g. JPEG quality).

imdecode(buf::AbstractVector{UInt8}, [flags])

Decode an encoded image held in a flat byte buffer (e.g. from imencode, a file read, or a network response) into a Mat. This convenience method reshapes buf into the 3D InputArray the generated imdecode expects; without flags it uses the same default as imread (IMREAD_COLOR_BGR). See issue #58.