Core concepts

OpenCV.jl is a thin, near-mechanical wrapper over the OpenCV C++ library. Knowing a few conventions makes the whole API predictable.

Naming: `cv::` → `OpenCV.`

Every wrapped function, enum, and class keeps its C++ name with the cv:: namespace removed, reached through the OpenCV. prefix:

C++	Julia
`cv::imread(path)`	`OpenCV.imread(path)`
`cv::cvtColor(src, dst, cv::COLOR_RGB2GRAY)`	`OpenCV.cvtColor(src, OpenCV.COLOR_RGB2GRAY)`
`cv::ORB::create()`	`OpenCV.ORB_create()`
`cv::IMREAD_GRAYSCALE`	`OpenCV.IMREAD_GRAYSCALE`

Nested names (class statics, nested enums) flatten with _, so cv::ORB::create becomes OpenCV.ORB_create and cv::TermCriteria::COUNT becomes OpenCV.TermCriteria_COUNT. Nothing is exported, so the OpenCV. prefix is always required — this keeps the thousands of short names (blur, add, mean, …) from colliding with Base or other packages.

Output arguments work as in C++/Python: functions return their results rather than mutating an output buffer you pass in.

Images are `Mat`s with `(channels, cols, rows)` axes

The central type is Mat, a 3-dimensional AbstractArray that shares memory with an OpenCV cv::Mat. Mind the axis order: a cv::Mat is exposed with axes (channels, cols, rows), i.e. size(img) == (nchannels, width, height).

image	`size(img)`
grayscale `H×W`	`(1, W, H)`
colour `H×W` (BGR)	`(3, W, H)`
ORB descriptors, `N` keypoints	`(1, descriptorSize, N)`

This is OpenCV's row-major height × width × channels buffer reinterpreted in Julia's column-major order, so no copy or transpose happens at the boundary. Indexing is img[channel, col, row].

Interop with Julia arrays

Any AbstractArray{T,3} with T one of the supported element types is accepted wherever an image is expected (this Union is OpenCV.InputArray), so you can pass a plain Julia array directly and wrap one as a Mat with OpenCV.Mat(A):

julia> using OpenCV

julia> A = rand(UInt8, 3, 640, 480);          # (channels, cols, rows)

julia> size(OpenCV.Mat(A))                    # zero-copy view; shares memory with A
(3, 640, 480)

julia> size(OpenCV.cvtColor(A, OpenCV.COLOR_BGR2GRAY))   # plain arrays work too
(1, 640, 480)

The supported element types are UInt8, Int8, UInt16, Int16, Int32, Float32, and Float64 (OpenCV's CV_8U … CV_64F).

Scalars, points, and small vectors

A colour/scalar value is a plain tuple of up to four numbers, e.g. (0, 255, 255) for yellow; missing components default to 0.
Geometry primitives are immutable structs parameterised by element type: Point, Point3, Size, Rect, RotatedRect, Range, TermCriteria. Construct them with the element type to match the wrapped signature, e.g. OpenCV.Size{Int32}(640, 480).
Per-pixel multi-channel values returned from a Mat are Vecs.

Reading and writing images

Use the native OpenCV.imread/OpenCV.imwrite, or the FileIO interface (load/save) for BMP/JPEG/PNG/TIFF. See Reading and Writing of Images for details and flags. ```