Facial recognition in videos – Exploring Video Data

Concept: Facial recognition involves identifying and verifying faces in videos. It’s used in security systems, user authentication, and various human-computer interaction applications.

Tools: OpenCV (for face detection), Dlib (for facial landmark detection), and face recognition libraries (e.g., face_recognition)

Here’s a brief overview of the steps in the facial recognition code:

1. Load face detector and landmark predictor: Load a pre-trained face detector (dlib.get_frontal_face_detector()) and a facial landmark predictor (dlib.shape_predictor(‘shape_predictor_68_face_landmarks.dat’)).
2. Video capture: Open a video file (sample_video.mp4) using cv2.VideoCapture.
3. Face detection loop: Iterate through frames in the video.
4. Detect faces: Use the face detector to identify faces in each frame.
5. Facial landmark detection: For each detected face, use the facial landmark predictor to locate facial landmarks.
6. Draw facial landmarks: Draw circles in the positions of the detected facial landmarks on the frame.
7. Draw bounding box: Draw a green bounding box around each detected face on the frame.
8. Display facial recognition result: Show the frame with facial landmarks and bounding boxes in a window named Facial Recognition using cv2.imshow.
9. Exit condition: Break the loop upon pressing the Esc key (cv2.waitKey).
10. Cleanup: Release the video capture object and close all OpenCV windows.

This code showcases a basic facial recognition application where faces are detected in each frame, and facial landmarks are drawn for each detected face. The bounding box outlines the face, and circles highlight specific facial features:
from deepface import DeepFace
import cv2
# Load a sample image
img_path1 = ‘./PacktPublishing/DataLabeling/ch08/data/pic1.jpeg’
img_path2 = ‘./PacktPublishing/DataLabeling/ch08/data/pic2.jpeg’
img = cv2.imread(img_path)
# Perform facial recognition
result = DeepFace.verify(img1_path=img_path1, img2_path=img_path2)
# Display the result
print(“Are these faces the same person?
“, result[“verified”])
# Additional information
print(“Facial recognition result:”, result)

We get the output that follows:
Are these faces the same person?
True
Facial recognition result: {‘verified’: True, ‘distance’: 0.20667349278322178, ‘threshold’: 0.4, ‘model’: ‘VGG-Face’, ‘detector_backend’: ‘opencv’, ‘similarity_metric’: ‘cosine’, ‘facial_areas’: {‘img1’: {‘x’: 74, ‘y’: 50, ‘w’: 713, ‘h’: 713}, ‘img2’: {‘x’: 63, ‘y’: 8, ‘w’: 386, ‘h’: 386}}, ‘time’: 0.48}

Video compression techniques

Video compression reduces the file size of videos, making them more manageable for storage, transmission, and processing.

Some common techniques are as follows:

• Lossy compression: Sacrifices some quality for reduced file size (e.g., H.264, H.265)

Video streaming platforms such as YouTube utilize lossy compression (H.264) to efficiently transmit videos over the internet. The sacrifice in quality ensures smoother streaming experiences, faster loading times, and reduced data usage for users.

• Lossless compression: Maintains original quality but with less compression (e.g., Apple ProRes, FFV1)

In professional video editing workflows, where preserving the highest possible quality is crucial, lossless compression is employed. Formats such as Apple ProRes or FFV1 are used for storing and processing video files without compromising quality. This is common in film production, video editing studios, and for archival purposes.