TIEVisionFaceDetector

Declaration

TIEVisionFaceDetector = interface(TIEVisionBase)

Description

Finds faces using the YuNet neural network (OpenCV FaceDetectorYN).
Compared with Haar cascades (TIEVisionObjectsFinder), it finds tilted, partly turned and small faces more reliably, gives fewer false detections, and returns a score and five facial landmarks for each face (TIEVisionDetectedFace).
The model is built into the IEVision DLL, so no extra files are needed.

Faces from about 10 pixels wide up to the full image are found. Large images are scaled down to setMaxSize pixels for speed.
Faces tilted up to about 60 degrees are found, and their landmarks are corrected for the tilt. Enable setDetectRotated to also find sideways and upside down faces.

Use alignFace to get upright, uniformly cropped faces, for example to train and query TIEVisionFaceRecognizer.

Shortcut methods are available to straighten a photo by the eyes, crop to the main face and obscure faces.

24 bit RGB, 32 bit RGBA, 8 bit gray scale and 1 bit images are supported.

Note: The YuNet model is licensed under the MIT license (Copyright © 2020 Shiqi Yu). See OpenCV Zoo.

Demos

Example

Also see: Automated Samples

var
  detector: TIEVisionFaceDetector;
  faces: TIEVisionVectorObjRef;
  face: TIEVisionDetectedFace;
  r: TIEVisionRect;
  i: Integer;
begin
  detector := IEVisionLib.createFaceDetector();
  faces := detector.detect( ImageEnView1.IEBitmap.GetIEVisionImage(), IEVisionRect( 0, 0, 0, 0 ));
  for i := 0 to faces.size() - 1 do
  begin
    face := TIEVisionDetectedFace( faces.getObj( i ));
    r := face.getRect();
    ImageEnView1.LayersAdd( iesRectangle, Rect( r.x, r.y, r.x + r.width, r.y + r.height ), clRed, 2 );
  end;
end;

Methods and Properties

See Also

◼createFaceDetector
◼TIEVisionDetectedFace
◼TIEVisionObjectsFinder
◼TIEVisionFaceRecognizer
◼DetectFaces
◼LevelByEyes
◼CropToFace
◼BlurFaces
◼detect

---