This application implements an image filter for smoothing the skin of a face. There are three heuristics which can be used for skin detection:

• Mean color detects the skin by calculating the mean color of the face region.

• Dominant color detects the skin by finding the dominant hue and saturation values in the face region

• Selective sampling detects the skin by sampling colors from the forehead, chin, and cheeks regions. It may be more accurate than other heuristics, but works well only for appropriately detected frontal faces with typical proportions.

It is assumed that OpenCV 4.x, C++17 compiler, and cmake 2.18.12 or newer are installed on the system.

Open CMakeLists.txt and set the correct OpenCV directory in the following line:

SET(OpenCV_DIR /home/hp/workfolder/OpenCV-Installation/installation/OpenCV-master/lib/cmake/opencv4)

Depending on the platform and the way OpenCV was installed, it may be needed to provide the path to cmake files explicitly. On my KUbuntu 20.04 after building OpenCV 4.4.0 from sources the working OpenCV_DIR looks like /lib/cmake/opencv4. On Windows 8.1 after installing a binary distribution of OpenCV 4.2.0 it is C:\OpenCV\build.

In the root of the project folder create the build directory unless it already exists. Then from the terminal run the following:

cd build
cmake ..

This should generate the build files. When it's done, compile the code:

cmake --build . --config Release

Create the models directory next to the executable file.

Download opencv_face_detector.pbtxt and opencv_face_detector_uint8.pb files to the models directory.

Refer to this page for model details.

The program has to be run from the command line. It takes in the image file path and several optional parameters:

skinsoft -image=<image file> [-confidence=<float 0..1>] [-radius=<integer>] [-sigmac=<float>] [-sigmas=<float>] [-heuristic=<choice>] [-help]

Sample usage (linux):

./skinsoft -image=../images/old_couple_1.jpg -confidence=0.5 -radius=4

To work correctly the program requires the face detection model files. Place the models folder, containing the .pbtxt and .pb files, in the directory from where the program is launched.