/// <summary>
/// Detects features on a grayscale image.
/// </summary>
/// <param name="img"></param>
/// <param name="storage"></param>
/// <returns></returns>
protected override List <Face> DetectFeatures(IplImage img, CvMemStorage storage)
{
//Determine minimum face size
var minSize = (int)Math.Round((double)MinSizePercent / 100.0 * Math.Min(img.Width, img.Height));
//Detect faces (frontal). TODO: side
Stopwatch watch = Stopwatch.StartNew();
CvAvgComp[] faces = Cv.HaarDetectObjects(img, Cascades["FaceCascade"], storage, 1.0850, MinConfidenceLevel, 0, new CvSize(minSize, minSize)).ToArrayAndDispose();
watch.Stop();
Debug.WriteLine("Face detection time = " + watch.ElapsedMilliseconds);
//Sort by accuracy
Array.Sort <CvAvgComp>(faces, CompareByNeighbors);
//Convert into feature objects list
List <Face> features = new List <Face>(faces.Length);
foreach (CvAvgComp face in faces)
{
features.Add(new Face(PolygonMath.ScaleRect(face.Rect.ToRectangleF(), ExpandX, ExpandY), face.Neighbors));
}
//Unless we're below MinFaces, filter out the low confidence matches.
while (features.Count > MinFaces && features[features.Count - 1].Accuracy < ConfidenceLevelThreshold)
{
features.RemoveAt(features.Count - 1);
}
//Never return more than [MaxFaces]
return((features.Count > MaxFaces) ? features.GetRange(0, MaxFaces) : features);
}
/// <summary>
/// Detects features on a grayscale image.
/// </summary>
/// <param name="img"></param>
/// <param name="storage"></param>
/// <returns></returns>
protected override List <Face> DetectFeatures(IplImage img, CvMemStorage storage)
{
//Determine minimum face size
var minSize = Math.Max(12, (int)Math.Round((double)MinSizePercent / 100.0 * Math.Min(img.Width, img.Height)));
//Detect faces (frontal).
Stopwatch watch = Stopwatch.StartNew();
CvAvgComp[] faces = BorrowCascade("FaceCascadeAlt", c => Cv.HaarDetectObjects(img, c, storage, 1.0850, MinConfidenceLevel, HaarDetectionType.DoCannyPruning, new CvSize(minSize, minSize), new CvSize(0, 0)).ToArrayAndDispose());
//Sort by accuracy
Array.Sort <CvAvgComp>(faces, CompareByNeighbors);
//Convert into feature objects list
List <Face> features = new List <Face>(faces.Length);
foreach (CvAvgComp face in faces)
{
features.Add(new Face(PolygonMath.ScaleRect(face.Rect.ToRectangleF(), ExpandX, ExpandY), face.Neighbors));
}
// Doesn't add much, and would have to be deduplicated.
//CvAvgComp[] profiles = BorrowCascade("FaceProfile", c => Cv.HaarDetectObjects(img, c, storage, 1.2, MinConfidenceLevel + 2, HaarDetectionType.FindBiggestObject | HaarDetectionType.DoRoughSearch | HaarDetectionType.DoCannyPruning, new CvSize(img.Width / 8, img.Height / 8), new CvSize(0, 0)).ToArrayAndDispose());
//foreach (CvAvgComp face in profiles) features.Add(new Face(PolygonMath.ScaleRect(face.Rect.ToRectangleF(), ExpandX, ExpandY), face.Neighbors));
// Test for eyes, if faces > 20 pixels
foreach (var face in features)
{
var w = (int)(face.X2 - face.X);
var h = (int)((face.Y2 - face.Y) * 0.6);
if (w > 20)
{
img.SetROI((int)face.X, (int)face.Y, w, h);
storage.Clear();
CvAvgComp[] eyes = BorrowCascade("Eye",
c => Cv.HaarDetectObjects(img, c, storage, 1.0850, 4, HaarDetectionType.FindBiggestObject | HaarDetectionType.DoRoughSearch,
new CvSize(4, 4), new CvSize(img.Width / 2, img.Height / 2))
.ToArrayAndDispose());
if (eyes.Length == 0)
{
// Halve the estimated accuracy if there are no eyes detected
face.Accuracy = face.Accuracy / 2;
// We never want to boost accuracy, because the walls have eyes
}
}
}
//Unless we're below MinFaces, filter out the low confidence matches.
while (features.Count > MinFaces && features[features.Count - 1].Accuracy < ConfidenceLevelThreshold)
{
features.RemoveAt(features.Count - 1);
}
watch.Stop();
totalTime += watch.ElapsedMilliseconds;
count++;
Debug.WriteLine($"Face detection time: {watch.ElapsedMilliseconds}ms (avg {totalTime / count}ms)");
//Never return more than [MaxFaces]
return((features.Count > MaxFaces) ? features.GetRange(0, MaxFaces) : features);
}
