real-time-de-identification.py

import cv2
import numpy as np
import tensorflow as tf
from tensorflow import keras
from scipy.ndimage import gaussian_filter

# Color Code
RED = (0, 0, 255)
GREEN = (0, 255, 0)

# Instead of Streaming
VIDEO_FILE = '/Users/praharshgurudatta/Downloads/My Videos/202109080914_45_1.mp4'

# Detect Face
face_cascade = cv2.CascadeClassifier(
    cv2.data.haarcascades + 'haarcascade_frontalface_default.xml')

# Load Model
model_name = 'face_classifier.h5'
face_classifier = keras.models.load_model(f'models/{model_name}')
class_names = ['me', 'not_me']


# Check the Coordinates will be Non-negative
def get_extended_image(img, x, y, w, h, k=0.1):  # img, x, y, w, h: from video
    if x - k * w > 0:
        start_x = int(x - k * w)
    else:
        start_x = x
    if y - k * h > 0:
        start_y = int(y - k * h)
    else:
        start_y = y

    end_x = int(x + (1 + k) * w)
    end_y = int(y + (1 + k) * h)

    face_image = img[start_y:end_y, start_x:end_x]
    face_image = tf.image.resize(face_image, [250, 250])  # Check the Shape
    # shape from (250, 250, 3) to (1, 250, 250, 3)
    face_image = np.expand_dims(face_image, axis=0)
    return face_image


# Streaming or Video File
video_capture = cv2.VideoCapture()

if not video_capture.isOpened():
    print("Unable to access the camera, Replaces with saved video files")
    video_capture = cv2.VideoCapture(VIDEO_FILE)  # Instead of Streaming
else:
    print("Access to the camera was successfully obtained")
    video_capture = cv2.VideoCapture(0)  # Check the Activated Cam Number

print("Streaming started - to quit press ESC")


# Loop for Real-Time
while True:

    ret, frame = video_capture.read()
    if not ret:
        print("Can't Receive Frame, Exit Streaming")
        break

    gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)

    faces = face_cascade.detectMultiScale(
        gray,
        scaleFactor=1.3,
        minNeighbors=5,
        minSize=(100, 100),
        flags=cv2.CASCADE_SCALE_IMAGE
    )

    # Loop
    for (x, y, w, h) in faces:
        face_image = get_extended_image(frame, x, y, w, h, 0.5)

        # Predict Result
        result = face_classifier.predict(face_image)
        prediction = class_names[np.array(
            result[0]).argmax(axis=0)]  # predicted class
        confidence = np.array(result[0]).max(axis=0)  # degree of confidence

        # Draw a rectangle around the face
        if prediction == 'me':
            # <de-identification>
            # Select One Option You Want, Then Delete the Other Options

            # Opt 1. gaussian_filter() : face_blurred
            # face_blurred = frame
            # face_blurred[y:y + h, x:x + w] = gaussian_filter(face_blurred[y:y + h, x:x + w], 10)

            # Opt 2. cv2.blur() : cv2_blur
            # cv2_blur = frame
            # cv2_blur[y:y + h, x:x + w] = cv2.blur(cv2_blur[y:y + h, x:x + w], (30, 30))

            # Opt 3. cv2.GaussianBlur() : cv2_Gau_blur
            cv2_Gau_blur = frame
            cv2_Gau_blur[y:y + h, x:x + w] = cv2.GaussianBlur(cv2_Gau_blur[y:y + h, x:x + w], (9, 9), 10)

            # # Opt 4. Resize the image : mosaic
            # if len(faces):
            #     for (x, y, w, h) in faces:
            #         mosaic = frame[y:y + h, x:x + w]
            #         mosaic = cv2.resize(mosaic, dsize=(0, 0), fx=0.04, fy=0.04)  # Reduction
            #         mosaic = cv2.resize(mosaic, (w, h), interpolation=cv2.INTER_AREA)  # Expansion
            #         frame[y:y + h, x:x + w] = mosaic

            # color = GREEN

            # Insert the Option You Choose
            cv2.rectangle(mosaic,  # Original: frame
                          (x, y),
                          (x + w, y + h),
                          color,
                          2)  # thickness in px
            cv2.putText(frame,
                        # text to put
                        "{:6} - {:.2f}%".format(prediction, confidence * 100),
                        (x, y),
                        cv2.FONT_HERSHEY_PLAIN,  # font
                        2,  # fontScale
                        color,
                        2)  # thickness in px
        else:
            color = RED
            # Draw a rectangle around the face
            cv2.rectangle(frame,
                          (x, y),
                          (x + w, y + h),
                          color,
                          2)  # thickness in px
            cv2.putText(frame,
                        # text to put
                        "{:6} - {:.2f}%".format(prediction, confidence * 100),
                        (x, y),
                        cv2.FONT_HERSHEY_PLAIN,  # font
                        2,  # fontScale
                        color,
                        2)  # thickness in px

    # Display
    cv2.imshow("Real-time Face detector", frame)

    # Exit with ESC
    key = cv2.waitKey(1)
    if key % 256 == 27:  # ESC code
        break

video_capture.release()
cv2.destroyAllWindows()