suszw / opencv-ndarray-conversion

NumPy ndarray ⇋ OpenCV Mat conversion, that just works.

NumPy ndarray ⇋ OpenCV Mat conversion, that just works.

API

• class NDArrayConverter: The converter class

• NDArrayConverter::NDArrayConverter(): Constructor

• cv::Mat NDArrayConverter::toMat(const PyObject* o): Convert a NumPy ndarray to a cv::Mat.

  • o is the object representing the Python representation of the ndarray.

  • Returns a cv::Mat which is the OpenCV representation of o.

• PyObject* NDArrayConverter::toNDArray(const cv::Mat& mat): Convert a cv::Mat to a NumPy ndarray.

  • mat is the cv::Mat to convert.

  • Returns a PyObject* that is the Python representation of an ndarray.

Examples

examples.cpp contains

• An implementation of matrix multiplication mul() that takes two ndarray objects, converts them to cv::Mat, multiplies them and returns the result as an ndarray.

• An image binarization function binarize() that takes an ndarray containing a grayscale image and a threshold value, converts the image to a cv::Mat and thresholds(binarizes) it. It then returns the result as a ndarray.

• An image display function display(ndarray) that just takes any image as an ndarray object and displays it.

All of these functions are callable from Python.

The plumbing(passing of values to and from Python) is handled by Boost::Python.

test.py contains some testcases that call the aforementioned C++ functions from Python.

Building and testing

After installing Boost::Python and NumPy(maybe the devel package),

$ make
$ make test

This builds and tests the project. Specifically, it builds the core converter in conversion.o and a set of examples in examples.so, which is a Python module containing a few examples of using the converter. test.py contains a test-suite to test (and demonstrate) the converter.