Drawing Bounding Boxes
Best practices for drawing and labelling detection results on images.
6 min•By Priygop Team•Updated 2026
Tip
Tip
If the Haar cascade is detecting too many false positives (non-faces detected as faces), increase minNeighbors (try 6 or 8). If it is missing real faces, decrease minNeighbors (try 3 or 4) or decrease scaleFactor (try 1.05). Also try equalising the histogram before detection — cv2.equalizeHist(gray) often significantly improves accuracy in low-light or high-contrast images.
Diagram
Loading diagram…
Machine Learning follows a structured pipeline from data to deployment