人脸关键点检测及人脸分割

人脸关键点检测，输入opencv格式的bgr通道图，默认在输入图上画关键点，也可以在空白图上画，如果要分割出人脸，可以使用矩形框框住关键点，然后在原图上裁剪，下面是一个成功案例。

源代码

import mediapipe as mp
import numpy as np
# 输入bgr通道图，输出人脸网点图
# mode=1在原图上画，mode=2在空白图上画
def get_face(image,mode=1):
    mp_drawing = mp.solutions.drawing_utils
    mp_face_mesh = mp.solutions.face_mesh
    drawing_spec = mp_drawing.DrawingSpec(thickness=1, circle_radius=1, color=(0, 255, 0))#bgr
    with mp_face_mesh.FaceMesh(
        max_num_faces=2,
        min_detection_confidence=0.5,
        min_tracking_confidence=0.5) as face_mesh:
        image = cv2.cvtColor(cv2.flip(image, 1), cv2.COLOR_BGR2RGB)
        # To improve performance, optionally mark the image as not writeable to
        # pass by reference.
        image.flags.writeable = False
        results = face_mesh.process(image)
 
        # Draw the face mesh annotations on the image.
        # and print landmarks' id, x, y, z 
        image.flags.writeable = True
        image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
        if mode==1:
            mask=image
          
        elif mode==2:
            #新建空白图
            mask=np.zeros(image.shape,dtype=np.uint8)
        if results.multi_face_landmarks:
            for face_landmarks in results.multi_face_landmarks:
                # Draw landmarks on the image.
                mp_drawing.draw_landmarks(
                    image=mask,#在空白图上画
                    landmark_list=face_landmarks,
                    connections=mp_face_mesh.FACEMESH_CONTOURS,
                    landmark_drawing_spec=drawing_spec,
                    connection_drawing_spec=drawing_spec)
                # print id, x, y, z
                # time cost 
                if mode==2:
                    xs=[]
                    ys=[]
                    zs=[]
                    for id,lm in enumerate(face_landmarks.landmark):
                        ih, iw, ic = image.shape
                        x,y = int(lm.x*iw), int(lm.y*ih)
                        xs.append(x)
                        ys.append(y)
                        zs.append(lm.z)
                        # print(id, x,y,lm.z)
                    #用矩形框住这些点
                    box=cv2.boundingRect(np.array([xs,ys]).T)
                    #裁剪原图
                    face=mask[box[1]:box[1]+box[3],box[0]:box[0]+box[2]]
                    #保持比例缩放到128*128
                    return face
                elif mode==1:
                    return mask
    #没有人脸              
    if mode==1:  
        return image
    elif mode==2:
        return np.zeros(image.shape,dtype=np.uint8)