使用 OpenCV 和Python模糊和匿名人脸

在本文中，我们将了解如何使用 OpenCV 和Python对人脸进行模糊处理和匿名化处理。

为此，我们将使用级联分类器来检测人脸。确保从以下链接下载相同的内容：haarcascade_frontalface_default.xml

方法

• 首先，我们使用内置的人脸检测算法，从实时视频或图像中检测人脸。在这里，我们将使用级联分类器方法从实时视频中检测人脸（使用网络摄像头）。
• 然后，读取来自实时视频的帧。存储最新帧并转换为灰度，以更好地了解特征。
• 现在，为了使输出美观，我们将在检测到的人脸周围制作一个带颜色边框的矩形。但是，我们希望检测到的人脸是模糊的，所以我们使用 medianBlur函数来做同样的事情，并提到人脸应该模糊的区域。
• 而且，现在我们要显示模糊的脸，使用 imshow函数读取的帧，我们希望它显示出来，直到我们按下一个键。

逐步实施：

第 1 步：导入人脸检测算法，称为级联分类器。

Python3

import cv2
  
# to detect the face of the human
cascade = cv2.CascadeClassifier("haarcascade_frontalface_default.xml")

第 2 步：从视频中捕获帧，以便从帧中检测人脸

Python3

video_capture = cv2.VideoCapture(0)
while True:
    
    # capture the latest frame from the video
    check, frame = video_capture.read()

第 3 步：将捕获的帧更改为灰度。

Python3

# convert the frame into grayscale(shades of black & white)
gray_image = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
face = cascade.detectMultiScale(gray_image,
                                scaleFactor=2.0,
                                minNeighbors=4)

第 4 步：在检测到的人脸周围绘制一个彩色矩形。

Python3

for x, y, w, h in face:
  
    # draw a border around the detected face.
    # (here border color = green, and thickness = 3)
    image = cv2.rectangle(frame, (x, y),
                          (x+w, y+h), 
                          (0, 255, 0), 3)

第5步：模糊矩形内的部分（包含检测到的人脸）。

Python3

# blur the face which is in the rectangle
image[y:y+h, x:x+w] = cv2.medianBlur(image[y:y+h, x:x+w], 35)

步骤6：显示最终输出，即检测到的人脸（在矩形内）是模糊的。

Python3

# show the blurred face in the video
cv2.imshow('face blurred', frame)
key = cv2.waitKey(1)

下面是完整的实现：

Python3

import cv2
  
# to detect the face of the human
cascade = cv2.CascadeClassifier("haarcascade_frontalface_default.xml")
  
# VideoCapture is a function, to capture
# video from the camera attached to system
# You can pass either 0 or 1
# 0 for laptop webcam
# 1 for external webcam
video_capture = cv2.VideoCapture(0)
  
# a while loop to run infinite times,
# to capture infinite number of frames for video
# because a video is a combination of frames
while True:
    
    # capture the latest frame from the video
    check, frame = video_capture.read()
  
    # convert the frame into grayscale(shades of black & white)
    gray_image = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
  
    # detect multiple faces in a captured frame
    # scaleFactor: Parameter specify how much the
    # image sizeis reduced at each image scale.
    # minNeighbors: Parameter specify how many
    # neighbours each rectangle should have to retain it.
    # rectangle consists the detect object.
    # Here the object is the face.
    face = cascade.detectMultiScale(
        gray_image, scaleFactor=2.0, minNeighbors=4)
  
    for x, y, w, h in face:
  
        # draw a border around the detected face. 
        # (here border color = green, and thickness = 3)
        image = cv2.rectangle(frame, (x, y), (x+w, y+h), 
                              (0, 255, 0), 3)
  
        # blur the face which is in the rectangle
        image[y:y+h, x:x+w] = cv2.medianBlur(image[y:y+h, x:x+w],
                                             35)
  
    # show the blurred face in the video
    cv2.imshow('face blurred', frame)
    key = cv2.waitKey(1)
  
    # This statement just runs once per frame.
    # Basically, if we get a key, and that key is a q,
    if key == ord('q'):
        break
  
# we will exit the while loop with a break,
# which then runs:
video_capture.release()
cv2.destroyAllWindows()

输出：