System information (version)

• OpenCV => 4.5.4

Detailed description

This is proposal. Let's discuss.

• Move cv::FaceDetectorYN/cv::FaceRecgnizer to DNN module from ObjDetect module
  Currently, cv::FaceDetectorYN/cv::FaceRecgnizerSF is implemented in the ObjDetect module.
  cv::FaceDetectorYN/cv::FaceRecgnizer is a high-level API for face detection and recognition.
  So, I think it is natural to have it in the DNN module. I suggest move this.

• Re-implementation to high-level API extended from cv::dnn::Model
  All high-level APIs in DNN modules inherit from cv::dnn::Model.
  cv::FaceDetectorYN/cv::FaceRecgnizerSF should also be re-implemented to inherit from cv::dnn::Model.
  It is different post processing depending on models. For example, YuNet and OpenCV Face Detector.
  It is better to implement like cv::dnn::TextDetectionModel->cv::dnn::TextDetectionModel_DB and cv::dnn::TextDetectionModel->cv::dnn::TextDetectionModel_EAST, as in text detection model.

• Support for OpenCV Face Detector
  It should add a high-level API that supports the traditional OpenCV Face Detector.
  Until now, users have been using the cv::dnn::DetectionModel to using OpenCV Face Detector with high-level API.
  There was no hight-level API for face detection. I think we should take this opportunity to create one.
  cv::dnn::FaceDetectionModel->cv::dnn::FaceDetectionModel_OpenCV (Alternatively, it could be named cv::dnn::FaceDetectionModel_SSD because OpenCV Face Detector is SSD.) and cv::dnn::FaceDetectionModel->cv::dnn::FaceDetectionModel_YN.

• Change the output format of cv::FaceDetectorYN
  Currently, cv::FaceDetectorYN outputs the bounding box, landmarks, and confidence score in a single Mat. It is not user friendly.
  For example, in the cv::dnn::ObjectDetectionModel, the bounding box, class id, and confidence score are output in the respective Mat.
  cv::FaceDetectorYN should also output the bounding box, landmarks, and confidence score to each Mat.