Face Detection with OpenCV and Dlib in Python

This guide demonstrates how to build a face‑detection pipeline in Python using OpenCV and Dlib, covering three main approaches: Haar cascade classifiers, Dlib's Histogram of Oriented Gradients (HOG) detector, and Dlib's convolutional neural network (CNN) detector.

Installation

First install the required libraries:

pip install opencv-python
pip install dlib

Locate the pre‑trained model files (paths may vary by Python version):

/usr/local/lib/python3.7/site-packages/cv2/data

Haar Cascade Detection

Load the cascade XML files and create classifiers:

cascPath = "/usr/local/lib/python3.7/site-packages/cv2/data/haarcascade_frontalface_default.xml"
eyePath = "/usr/local/lib/python3.7/site-packages/cv2/data/haarcascade_eye.xml"
smilePath = "/usr/local/lib/python3.7/site-packages/cv2/data/haarcascade_smile.xml"
faceCascade = cv2.CascadeClassifier(cascPath)
eyeCascade = cv2.CascadeClassifier(eyePath)
smileCascade = cv2.CascadeClassifier(smilePath)

Detect faces in a grayscale image and draw rectangles:

faces = faceCascade.detectMultiScale(gray, scaleFactor=1.1, minNeighbors=5, flags=cv2.CASCADE_SCALE_IMAGE)
for (x, y, w, h) in faces:
    cv2.rectangle(gray, (x, y), (x+w, y+h), (255, 255, 255), 3)

The same pattern is used for eyes and smiles, adjusting scaleFactor and minNeighbors as needed.

Haar Feature Theory

Haar features are simple rectangular intensity differences that can be computed efficiently with integral images, allowing rapid evaluation of millions of candidate windows.

Dlib HOG Detector

Initialize the HOG‑based detector and apply it to a grayscale frame:

face_detect = dlib.get_frontal_face_detector()
rects = face_detect(gray, 1)
for rect in rects:
    (x, y, w, h) = face_utils.rect_to_bb(rect)
    cv2.rectangle(gray, (x, y), (x+w, y+h), (255, 255, 255), 3)

The HOG approach is fast and works well for moderate‑size faces.

Dlib CNN Detector

Download the pre‑trained CNN model ( mmod_human_face_detector.dat ) and load it:

dnnFaceDetector = dlib.cnn_face_detection_model_v1("mmod_human_face_detector.dat")
rects = dnnFaceDetector(gray, 1)
for rect in rects:
    x1 = rect.rect.left()
    y1 = rect.rect.top()
    x2 = rect.rect.right()
    y2 = rect.rect.bottom()
    cv2.rectangle(gray, (x1, y1), (x2, y2), (255, 255, 255), 3)

This CNN provides the highest accuracy, especially for small or occluded faces, at the cost of slower processing.

Real‑Time Video Detection

Capture video from the webcam, convert each frame to grayscale, run the chosen detector, draw bounding boxes, and display the result. Press q to quit:

video_capture = cv2.VideoCapture(0)
while True:
    ret, frame = video_capture.read()
    gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
    rects = faceCascade.detectMultiScale(gray, scaleFactor=1.1, minNeighbors=5, flags=cv2.CASCADE_SCALE_IMAGE)
    for (x, y, w, h) in rects:
        cv2.rectangle(frame, (x, y), (x+w, y+h), (0, 255, 0), 2)
    cv2.imshow('Video', frame)
    if cv2.waitKey(1) & 0xFF == ord('q'):
        break
video_capture.release()
cv2.destroyAllWindows()

Model Selection

When speed is critical, the HOG detector is the fastest; the Haar cascade offers comparable speed with slightly lower accuracy, while the CNN yields the best detection rates but is slower. Choose based on your real‑time requirements.