I recently read an article regarding facial recognition algorithms and how they make it easy to identify individuals from a single photo. But with the help of advances in machine learning, it’s also easier than ever to anonymize photos and videos. Based on this, a London AI startup had built an app with a focus on investigative journalists who wanted an easy way to record anonymous footage.
I was well aware of OpenCV but never applied/used it myself so thought this could be an interesting little project to see if I could replicate the functionality of this app directly in Python. So after some poking around the documentation and some experimentation later come up with the following simple POC
# Importing openCV and creating the classifier (included three but face1.xml performed best on my personal trials)
import cv2
classifier = cv2.CascadeClassifier('classifiers/face1.xml')
# Setting the webcam and resolutions
video = cv2.VideoCapture(0)
out = cv2.VideoWriter('face_blur.mp4', -1, 20.0, (640,480))
# Loop to keep the live stream running / detecting regions of interest
while True:
try:
ret, frames = video.read()
gray = cv2.cvtColor(frames, cv2.COLOR_BGR2GRAY)
# Here we now initiate the face detection and bounding box
faces = classifier.detectMultiScale(gray)
for (x, y, w, h) in faces:
cv2.rectangle(frames, (x, y), (x + w, y + h), (0, 0, 0), 2)
# Here we set the gaussian blur within the box
face_color = frames[y:y + h, x:x + w]
blur = cv2.GaussianBlur(face_color, (75, 75), 0)
frames[y:y + h, x:x + w] = blur
# Setting the video and OpenCV window
out.write(frames)
cv2.imshow('Face Blurring', frames)
if cv2.waitKey(1) & 0xFF == ord('q'): break
except Exception as e:
print(str(e))
break
# Turn off the camera and close the OpenCV windows
video.release()
out.release()
cv2.destroyAllWindows()
Below you can see a small snippet of the outcome - genuinely the performance impressed me massively for an out the box pre-trained model… I think it goes to show again how far the democratization of technology has come - a few years ago it would have been unforeseeable to create something like this on a personal laptop while watching a football match & beer (#multitasking!) - was an interesting way to spend a rainy afternoon, if your interested you can grab my code and classifiers over here
