首先加载数据集:
trump = cv2.imread('C:\\Users\\shm\\Desktop\\python code\\trump.jpg')
xml文件是一个人脸数据的级联分类器,给人脸数据,可以返回识别人脸的对象
分类器来源:https://github.com/opencv/opencv/tree/master/data/haarcascades
detector = cv2.CascadeClassifier('C:\\Users\\shm\\Desktop\\python code\\opencv_classifier\\haarcascade_frontalface_default.xml')
face_zone = detector.detectMultiScale(trump,scaleFactor = 1.3,minNeighbors = 5)
#默认值是scaleFactor = 1.1,minNeighbors = 3,这两个参数值越大限制条件越严格
#minsize,maxsize = (m,n)规定识别的人脸最大最小值,超过这个范围不会识别。
print(face_zone)
for x,y,z,n in face_zone:
cv2.rectangle(trump , pt1 = (x,y),pt2 = (x+z,y+n),color = \[0,0,255\],thickness = 2)#画一个正方形框框
cv2.imshow('trump',trump)
cv2.waitKey(0)
cv2.destroyAllWindows()
cv2.imwrite('C:\\Users\\shm\\Desktop\\python_data\\image_data\\trump.jpg',trump)
多个人脸时:
photo = cv2.imread('C:\\Users\\shm\\Desktop\\python code\\photo.jpg')
detector = cv2.CascadeClassifier('C:\\Users\\shm\\Desktop\\python code\\opencv_classifier\\haarcascade_frontalface_default.xml')
face_zone = detector.detectMultiScale(photo,scaleFactor = 1.3,minNeighbors = 5)
#默认值是scaleFactor = 1.1,minNeighbors = 3,这两个参数值越大限制条件越严格
#minsize,maxsize = (m,n)规定识别的人脸最大最小值,超过这个范围不会识别。
print(face_zone)
for x,y,z,n in face_zone:
cv2.rectangle(photo , pt1 = (x,y),pt2 = (x+z,y+n),color = [0,0,255],thickness = 2)#画一个正方形框框
#也可以画圆形框:cv2.circle(photo,center =(x+z//2,y+n//2),redius = z//2,color = [0,0,255])
cv2.imshow('trump',photo)
cv2.waitKey(0)
cv2.destroyAllWindows()
cv2.imwrite('C:\\Users\\shm\\Desktop\\python_data\\image_data\\photo.jpg',photo)
总结:对于上面的多人脸图片并不能提取全部人脸,
因此对于一些多人脸图片或人脸比较小的图片,应当边调参边训练,找到最优参数。