Connected Components (with CUDA)

The connected components problem is common in image processing. FolkeV's repository is an extension of original Playne's code with the Playne-Equivalence Connected-Component Labelling Algorithm described in:

D. P. Playne and K. Hawick,
"A New Algorithm for Parallel Connected-Component Labelling on GPUs,"
in IEEE Transactions on Parallel and Distributed Systems,
vol. 29, no. 6, pp. 1217-1230, 1 June 2018.

• URL: https://ieeexplore.ieee.org/document/8274991

    

Here, the FolkeV's code was download in the CUDA_CCL folder to be used as library with some other features (area and coordinates extraction) very common in OpenCV C++ libraries but not available in cv::cuda classes to be exectued on the GPU.

Usage

• Declare cudalabel class object:

cudalabel labels;

• Set image from CPU:

labels.setimg(image);

• Or, alternatively, set from GPU to avoid internal uploads, depending on your code needs:

labels.setgpuimg(gpuimage);

• Step (2), preprocess (uses FolkeV's code in the CPU version).

labels.preprocess();

• Step (3), internal call to the Playne&Hawick (2018) cuda code.

labels.labelize();

• Step (4), new functions, get the labels coordinates (internal call to kernel kgetinfo):

unsigned int** labelinfo;
labelinfo = labels.getinfo();

• Release resources (required if work continues).

labels.reset();

• Optional (generate image and save results):

if (labels.imgen())
    labels.lsave(output_name);  

Compilation

Download code and call cmake.

$ git clone https://github.com/armengot/cudalabel
$ cd cudalabel
$ mkdir build
$ cd build
$ cmake ..
$ make

CPU/GPU times

Using timing executable tool. Using the CPU image.

[CPU] processing image [2880x2160] with ../samples/sample0.jpg took 59 milliseconds.
[CPU] processing image [2880x2160] with ../samples/sample1.jpg took 61 milliseconds.
[CPU] processing image [2880x2160] with ../samples/sample2.jpg took 61 milliseconds.
[CPU] processing image [2880x2160] with ../samples/sample3.jpg took 61 milliseconds.
[CPU] processing image [2880x2160] with ../samples/sample4.jpg took 60 milliseconds.
[CPU] processing image [2880x2160] with ../samples/sample5.jpg took 62 milliseconds.
[CPU] processing image [2880x2160] with ../samples/sample6.jpg took 62 milliseconds.
[CPU] processing image [2880x2160] with ../samples/sample7.jpg took 62 milliseconds.
[CPU] processing image [2880x2160] with ../samples/sample8.jpg took 61 milliseconds.
[CPU] processing image [2880x2160] with ../samples/sample9.jpg took 62 milliseconds.

Uploading the image to the GPU.

[GPU] processing image [2880x2160] with ../samples/sample0.jpg took 129 milliseconds.
[GPU] processing image [2880x2160] with ../samples/sample1.jpg took 46  milliseconds.
[GPU] processing image [2880x2160] with ../samples/sample2.jpg took 45  milliseconds.
[GPU] processing image [2880x2160] with ../samples/sample3.jpg took 46  milliseconds.
[GPU] processing image [2880x2160] with ../samples/sample4.jpg took 47  milliseconds.
[GPU] processing image [2880x2160] with ../samples/sample5.jpg took 47  milliseconds.
[GPU] processing image [2880x2160] with ../samples/sample6.jpg took 51  milliseconds.
[GPU] processing image [2880x2160] with ../samples/sample7.jpg took 48  milliseconds.
[GPU] processing image [2880x2160] with ../samples/sample8.jpg took 47  milliseconds.
[GPU] processing image [2880x2160] with ../samples/sample9.jpg took 49  milliseconds.