Native OpenCV Swift Framework





GitHub Stars



Last Commit

2yrs ago







3-clause BSD




LegoCV is native OpenCV framework built for Swift and Objective-C projects. It eliminates the need to use Objective-C++ and write bringing code, and allows for full compatibility with native Swift projects. The only dependency is the native OpenCV framework for iOS (and later macOS and tvOS).


Swift is one of the fastest evolving languages, but there is currently no way to use C++ frameworks directly, as it was possible be with Objective-C/C++.

This project's purpose is to create a simple, easy to use native Swift framework for OpenCV. The project adds Swift and Objective-C convenience methods, but translates to OpenCV API entirely.

The idea is to simply wrap OpenCV native C++ classes into lightweight Objective-C classes, which are then natively bridged to Swift, providing a thin layer on top of native OpenCV. Realm and EmguCV in C# use similar framework structure. Possibility for fully native cross-platform Swift version of OpenCV exists in the future.


The following examples display the difference with using LegoCV in Swift or Objective-C compared to vanilla OpenCV in C++. The example is extracted from Face detection sample code, included with LegoCV. On iOS it uses OCVVideoCamera wrapper class to get image stream from camera (wraps OpenCV's CvVideoCamera, to keep backward compatibility).

Swift (LegoCV):

let faceDetector = OCVCascadeClassifier();
faceDetector.load(path: "haarcascade_frontalface_alt2.xml")

func process(image: OCVMat) {
    let scale = 2.0

    let minSize = OCVSize(width: 30, height: 30)
    let size = CGSize(width: 140.0, height: 140.00).ocvSize
    let gray = OCVMat()
    let smallImage = OCVMat(rows: Int(round(Double(image.rows) / scale)), cols: Int(round(Double(image.cols) / scale)), type: .cv8U, channels: 1)
    // OpenCV Default Syntax requires to predefine both input and output
    OCVOperation.convertColor(from: image, to: gray, with: .BGR2GRAY)
    // LegoCV syntactic sugar allows you to perform operations directly on the input, only defining output.
    image.convertColor(to: gray, with: .BGR2GRAY)
    let grayImg = image.convertColor(with: .BGR2GRAY)
    OCVOperation.convertColor(from: image, to: gray, with: .BGR2GRAY)
    OCVOperation.resize(from: gray, to: smallImage, size: smallImage.size, fx: 0, fy: 0, interpolation: .linear)
    OCVOperation.equalizeHistogram(from: smallImage, to: smallImage)
    // Faces are returned as OCVRect instances, so they are mapped in Swift, as they are structs.
    var faces : [OCVRect] = faceDetector.detectMultiscale(with: smallImage, scaleFactor: 1.1, minNeighbours: 2, flags: 0, minSize: minSize).map { $0.rect }
    // More LegoCV objective syntactic sugar
    let result : OCVCascadeClassifierResult = faceDetector.detectMultiscale(on: smallImage, with: OCVCascadeClassifierOptions.default)
    faces = result.objects

Objective-C (LegoCV with Objective-C):

- (void)setupClassifier {
    self.faceDetector = [[OCVCascadeClassifier alloc] init];
    [self.faceDetector loadPath:@"haarcascade_frontalface_alt2.xml"];

- (void)processImage:(OCVMat *)image {
    double scale = 2.0;

    OCVSize minSize;
    minSize.width = 30;
    minSize.height = 30;
    OCVMat* gray = [[OCVMat alloc] init];
    OCVMat* smallImage = [[OCVMat alloc] initWithRows:round(image.rows / scale) cols:round(image.cols / scale) type: OCVDepthTypeCv8U, channels: 1)
    [OCVOperation convertColorFromSource:image toDestination:gray with:OCVColorConversionTypeBGR2GRAY];
    [OCVOperation resizeFromSource:gray toDestination:smallImage size:smallImage.size fx:0 fy:0 interpolation:OCVInterpolationTypeLinear];
    [OCVOperation equalizeHistogramFromSource:smallImage toDestination:smallImage];
    // Faces are returned as OCVRectValue instances, which wrap OCVRect structs.
    NSArray<OCVRectValue *>* faces = [self.faceDetector detectMultiscaleWith:smallImage scaleFactor:1.1 minNeighbours:2 flags: 0 minSize:minSize];
    // Call the face detector classifier
    OCVCascadeClassifierResult* result = [self.faceDetector detectMultiScaleOnImage:smallImage withOptions:[OCVCascadeClassifierOptions defaultOptions]];

C++ (OpenCV):

using namespace cv;

void setup () {
    _faceDetector = new CascadeClassifier();

void processImage(cv::Mat img) {
    double scale = 2.0;
    Mat gray, smallImg( cvRound (img.rows/scale), cvRound(img.cols/scale), CV_8UC1 );
    cvtColor( img, gray, COLOR_BGR2GRAY );
    resize( gray, smallImg, smallImg.size(), 0, 0, INTER_LINEAR );
    equalizeHist( smallImg, smallImg );

    cv::Size minSize(30,30);

    vector<cv::Rect> faceRects;

    // Faces are returned in provided faceRects vector
    _faceDetector->detectMultiScale(smallImg, faceRects, 1.1, 2, 0, minSize);

More examples, including Swift playgrounds can be found in the sample project.


As this is a project in progress, documentation will be added to Wiki as development progresses.


There is a smaller performance impact compared to pure native C++ code of OpenCV, due to Objective-C messaging system. If you need a high performance code, it is still recommended to write the algorithm in C++ and add bridges to LegoCV or Objective-C.



For LegoCV you need cmake. Install it with brew install cmake and make sure you have Xcode Command Line tools installed. Trigger with xcode-select --install to check. Also make sure you use latest Xcode version and not Beta for master branch.


LegoCV can be installed with CocoaPods or Carthage. It's only dependency is OpenCV framework, which can be downloaded from their website.

pod 'LegoCV'

# Use only specific modules
pod 'LegoCV/Core'
pod 'LegoCV/VideoIO'

LegoCV supports iOS 8 and higher.

Example Project

  1. First clone the project:

git clone git@github.com:legoless/legocv.git cd legocv

  1. Initialize submodules
git submodule init
git submodule update
  1. Build opencv2.framework from git repository.
/usr/bin/python opencv/platforms/ios/build_framework.py ios --dynamic
  1. Open LegoCV.xcodeproj and build.


BSD license, respect OpenCV license as well.

Rate & Review

Great Documentation0
Easy to Use0
Highly Customizable0
Bleeding Edge0
Responsive Maintainers0
Poor Documentation0
Hard to Use0
Unwelcoming Community0