利用dlib81人脸关键点提取额头脸颊ROI

先上实验结果

'''
人脸关键点检测——dlib
1  dlib.get_frontal_face_detector()  获取人脸检测器
2  dlib.shape_predictor()  预测人脸关键点
'''
# import cv2
#上面的无法自动补全代码
import cv2.cv2 as cv2
import matplotlib.pyplot as plt
import dlib
import numpy as np


#读取图片
img = cv2.imread("6.png")

#调用人脸检测器
detector = dlib.get_frontal_face_detector()

#加载 预测关键点模型（68个关键点）
predictor = dlib.shape_predictor("shape_predictor_81_face_landmarks.dat")

#灰度化
gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)

#人脸检测
faces = detector(gray, 1)  # 1:将图片放大一倍，便于识别人脸; 0 原始图像

a = np.arange(81)
print("a:",a)
b = np.arange(81)

if (len(faces) != 0):
    # 找到脸颊区域
    for i in range(len(faces)):
        landmarks = np.matrix([[p.x, p.y] for p in predictor(img, faces[i]).parts()])
        for idx, point in enumerate(landmarks):
            # 68点的坐标
            pos = (point[0, 0], point[0, 1])
             # ROI：脸颊区域
            a = a.tolist()  # 将矩阵转换成列表。
            b = b.tolist()
            a[idx] = point[0, 0]
            b[idx] = point[0, 1]
            a = np.array(a)  # 将数据转化为矩阵
            b = np.array(b)

    # 额头区域
    x1 = int(a[76])
    y1 = int(b[19])
    x2 = int(a[73])
    y2 = int(b[73])
    if (x1 < 0): # 防止人脸转动，x1位置超出图像
        x1 = 0
    forehead_ROI = img[y2:y1, x1:x2]
    cv2.imshow("forehead_ROI",forehead_ROI)
    cv2.waitKey()

    # 脸颊区域
    x1 = int((a[2] + a[4]) / 2)
    y1 = int((b[2] + b[4]) / 2)
    y2 = int((b[2] + b[23]) / 2)
    x2 = int((a[12] + a[14]) / 2)
    if (x1 < 0):
        x1 = 0
    cheek_ROI = img[y2:y1, x1:x2]
    cv2.imshow("cheek_ROI",cheek_ROI)
    cv2.waitKey()
    # 检测整个人脸区域
    for k, d in enumerate(faces):
        [x1, x2, y1, y2] = [d.left(), d.right(), d.top(), d.bottom()]
    face_ROI = img[y1:y2, x1:x2]

    cv2.imshow("face_ROI",face_ROI)
    cv2.waitKey()

#显示整个效果
plt.imshow(img[:, :, ::-1]) # BGR 2 RGB
plt.title("81_face_landmarks")
plt.axis("off")
# cv2.imwrite("dlib_81_face_landmarks10.jpg", img)
plt.show()

'''
人脸关键点检测——dlib
1  dlib.get_frontal_face_detector()  获取人脸检测器
2  dlib.shape_predictor()  预测人脸关键点
'''
# import cv2
#上面的无法自动补全代码
import cv2.cv2 as cv2
import matplotlib.pyplot as plt
import dlib
import numpy as np

#读取图片
img = cv2.imread("6.png")

#调用人脸检测器
detector = dlib.get_frontal_face_detector()

#加载 预测关键点模型（68个关键点）
predictor = dlib.shape_predictor("shape_predictor_81_face_landmarks.dat")

#灰度化
gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)

#人脸检测
faces = detector(gray, 1)  # 1:将图片放大一倍，便于识别人脸; 0 原始图像

a = np.arange(81)
b = np.arange(81)

if (len(faces) != 0):
    # 找到脸颊区域
    for i in range(len(faces)):
        landmarks = np.matrix([[p.x, p.y] for p in predictor(img, faces[i]).parts()])
        for idx, point in enumerate(landmarks):
            # 68点的坐标
            pos = (point[0, 0], point[0, 1])
             # ROI：脸颊区域
            a = a.tolist()  # 将矩阵转换成列表。
            b = b.tolist()
            a[idx] = point[0, 0]
            b[idx] = point[0, 1]
            a = np.array(a)  # 将数据转化为矩阵
            b = np.array(b)

    # 额头区域
    x1 = int(a[76])
    y1 = int(b[19])
    x2 = int(a[73])
    y2 = int(b[73])
    if (x1 < 0): # 防止人脸转动，x1位置超出图像
        x1 = 0
    forehead_ROI = img[y2:y1, x1:x2]
    cv2.rectangle(img, (x1,y2), (x2,y1), (0,255,0), 1)

    # cv2.imshow("forehead_ROI",forehead_ROI)
    # cv2.waitKey()


    # 脸颊区域
    x1 = int((a[2] + a[4]) / 2)
    y1 = int((b[2] + b[4]) / 2)
    y2 = int((b[2] + b[23]) / 2)
    x2 = int((a[12] + a[14]) / 2)
    if (x1 < 0):
        x1 = 0
    cheek_ROI = img[y2:y1, x1:x2]
    cv2.rectangle(img, (x1,y2), (x2,y1), (0,255,0), 1)
    # cv2.imshow("cheek_ROI",cheek_ROI)
    # cv2.waitKey()

    # 检测整个人脸区域
    for k, d in enumerate(faces):
        [x1, x2, y1, y2] = [d.left(), d.right(), d.top(), d.bottom()]

    face_ROI = img[y1:y2, x1:x2]
    cv2.rectangle(img, (x1, y2), (x2, y1), (255, 0, 0), 1)  # 蓝色
    # cv2.imshow("face_ROI",face_ROI)
    # cv2.waitKey()
    
cv2.imshow("ROI",img)
cv2.waitKey()

这个ROI区域确定是根据关键点索引，比较麻烦。引入imutils库，更加简单

查看imutils库使用

查看 face_utils的使用