API Reference

This page contains links to the detailed API references for each module, alongside core wrapper structs.

📦 API Reference by Module

Select a module below to view all its classes, enums, and methods:

NOTE

OpenCV 5 Module Reorganization: If you're coming from OpenCV 4.x, note that the calib3d module has been split into three separate modules in OpenCV 5: Calib (camera calibration), Stereo (stereo correspondence), and Geometry (geometric algorithms). The features2d module has similarly been renamed to Features.

Core Module — Matrix class (Mat), CUDA execution contexts, arrays, and basic mathematical operations.
Imgproc (Image Processing) — Image filters (blur, dilate, erode), color conversions, resize, histograms, and contour extraction.
Dnn (Deep Learning) — Deep Neural Network loading, layers execution, forward passes, classification, text detection, and tokenizers.
Objdetect (Object Detection) — Face detectors (YuNet), barcode decoders, QR detectors, and cascades.
Imgcodecs (Image Codecs) — File reading/writing (Imread/Imwrite), memory encoding/decoding, and animations.
Features (Feature Detection) — 2D feature detectors/descriptors (SIFT, ORB, FAST, ALIKED) and matchers.
Video (Motion & Tracking) — Optical flow, Kalman filters, and motion tracking.
VideoIO (Video Input/Output) — Video file reading (VideoCapture) and writing (VideoWriter).
Ptcloud (Point Cloud) — 3D point cloud structures and geometry.
Stereo (Stereo Correspondence) — Stereo matching, disparity algorithms, and calibration.
Stitching (Image Stitching) — Panoramas, image matching, and warpers.
Calib (Camera Calibration) — Camera calibration matrix computations, distortions, and calibration grids.
Photo (Computational Photography) — Inpainting, denoising, HDR, and high-level photography processing.
Geometry (Geometric Algorithms) — Bounding boxes, convex hulls, and geometric calculations.
Flann (Fast Nearest Neighbors) — Fast approximate nearest neighbors index matching.
Highgui (GUI / Display) — Window creation, image showing (Cv2.Imshow), trackbars, and simple UI interaction.

📦 Core Classes & Structs

1. Mat

The primary CPU matrix class, matching C++ cv::Mat. Holds an unmanaged pointer to image data on the CPU heap.

Member	Type	Description
`Rows`	`int`	The number of rows (height) of the matrix.
`Cols`	`int`	The number of columns (width) of the matrix.
`Type()`	`int`	Returns the data type and channel layout as an integer constant (e.g. 0 for CV_8UC1, 16 for CV_8UC3).
`Empty()`	`bool`	Returns `true` if the matrix holds no data (0 rows or 0 columns).
`Dispose()`	`void`	Releases the underlying C++ memory heap immediately.

2. CudaGpuMat

The GPU matrix class, matching C++ cv::cuda::GpuMat. Holds an unmanaged pointer to device memory in GPU VRAM.

Member	Type	Description
`Rows`	`int`	Height of the matrix in GPU VRAM.
`Cols`	`int`	Width of the matrix in GPU VRAM.
`Upload(Mat cpuMat)`	`void`	Blocking Call: Copies data from CPU memory (`Mat`) to GPU device memory.
`Download(Mat cpuMat)`	`void`	Blocking Call: Copies data from GPU device memory back to CPU memory.
`Dispose()`	`void`	Deallocates the device pointer in GPU VRAM immediately.
`DefaultAllocator()`	`IntPtr`	Static Method: Returns the pointer to OpenCV's global default GPU memory allocator (required when instantiating `CudaGpuMat` to prevent Access Violations).

🎨 Enumerations & Structs

1. Mat Type Constants

OpenCV uses integer constants for matrix type identifiers. Common values:

Constant	Value	Description
`CV_8UC1`	`0`	8-bit unsigned integer, single channel (Grayscale)
`CV_8UC3`	`16`	8-bit unsigned integer, three channels (BGR)
`CV_32FC1`	`5`	32-bit float, single channel (Depth maps)
`CV_32FC3`	`21`	32-bit float, three channels

Pass these as raw int values to methods that accept a type parameter.

2. Size

A lightweight C# struct mapping dimensions:

csharp

public struct Size
{
    public int Width;
    public int Height;
}

3. Scalar

A 4-element double vector, mapping color thresholds or background values:

csharp

public struct Scalar
{
    public double V0;
    public double V1;
    public double V2;
    public double V3;
}

🛡️ Error Handling & Interop

1. OpenCVException

When a native OpenCV C++ call fails (e.g., passing invalid dimensions to a filter), the wrapper detects the failure code and calls ErrorHelper.CheckError().

The wrapper catches C++ exceptions at the interop boundary, parses them, and throws a managed OpenCVException in C#.
This prevents memory leaks and ensures C++ crashes do not kill the C# process.

2. P/Invoke Memory Boundary

mermaid

sequenceDiagram
    participant C# Application
    participant C# Wrapper (OpenCV5Sharp)
    participant C++ Interop (opencv5sharp_native)
    participant OpenCV Engine (opencv_world)

    C# Application->>C# Wrapper: new CudaGpuMat(h, w, type, allocator)
    C# Wrapper->>C++ Interop: cuda_GpuMat_New_0(h, w, type, allocator)
    Note over C++ Interop: Resolves allocator pointer
    C++ Interop->>OpenCV Engine: new cv::cuda::GpuMat(...)
    OpenCV Engine-->>C++ Interop: Returns instance pointer
    C++ Interop-->>C# Wrapper: Returns IntPtr (handle)
    C# Wrapper-->>C# Application: Returns CudaGpuMat instance

API Reference ​

📦 API Reference by Module ​

📦 Core Classes & Structs ​

1. Mat ​

2. CudaGpuMat ​

🎨 Enumerations & Structs ​

1. Mat Type Constants ​

2. Size ​

3. Scalar ​

🛡️ Error Handling & Interop ​

1. OpenCVException ​

2. P/Invoke Memory Boundary ​