face swap ,use shape_predictor_68_face_landmarks
我的运行环境，Python 3.11.4；其他版本应该也可以运行，我没测试 安装 dlib
pip install dlib
安装 opencv
pip install opencv-python
模型文件shape_predictor_68_face_landmarks.dat已包含目录下
python face_swap.py
直接上代码：
import os
import dlib
import cv2
import numpy as np
import numpy as np
import matplotlib.pyplot as plt
pwd_path = os.path.abspath(os.path.dirname(__file__))
source_img_path = os.path.join(pwd_path, '1.jpg')
target_img_path = os.path.join(pwd_path, '2.jpg')
plt.rcParams["font.sans-serif"] = "SimHei"
plt.rcParams["axes.unicode_minus"] = False
def detect_face_landmarks(image, predictor):
# 使用dlib的face detector（如HOG或CNN）检测人脸位置
face_detector = dlib.get_frontal_face_detector()
faces = face_detector(image)
if len(faces) < 1:
raise ValueError("Exactly one face should be present in the image.")
# 使用shape_predictor预测面部特征点
face_shape = predictor(image, faces[0])
return face_shape
def show_img_landmarks(image, landmarks):
# 显示图片和特征点
for idx, point in enumerate(landmarks.parts()):
# 68点的坐标
pos = (point.x, point.y)
print(idx+1,pos)
# 利用cv2.circle给每个特征点画一个圈，共68个
cv2.circle(image, pos, 5, color=(0, 255, 0))
# 利用cv2.putText输出1-68
font = cv2.FONT_HERSHEY_SIMPLEX
cv2.putText(image, str(idx+1), pos, font, 0.5, (0, 0, 255), 1,cv2.LINE_AA)
face_rect=landmarks.rect
cv2.rectangle(image,(face_rect.left(),face_rect.top()) ,(face_rect.right(),face_rect.bottom()), (0, 255, 0), 2)
return image
def get_face_mask(image_size, face_landmarks):
"""
获取人脸掩模，包含轮廓
:param image_size: 图片大小
:param face_landmarks: 68个特征点
:return: image_mask, 掩模图片
"""
mask = np.zeros(image_size, dtype=np.uint8)
points = np.concatenate([face_landmarks[0:17], face_landmarks[26:16:-1]])
cv2.fillPoly(img=mask, pts=[points], color=1)
return mask
def draw_convex_hull(im, points, color):
points = cv2.convexHull(points)
cv2.fillConvexPoly(im, points, color=color)
def get_face_mask2(sz, landmarks):
LEFT_EYE_POINTS = list(range(42, 48))
RIGHT_EYE_POINTS = list(range(36, 42))
LEFT_BROW_POINTS = list(range(22, 27))
RIGHT_BROW_POINTS = list(range(17, 22))
NOSE_POINTS = list(range(27, 35))
MOUTH_POINTS = list(range(48, 61))
OVERLAY_POINTS = [
LEFT_EYE_POINTS + RIGHT_EYE_POINTS + LEFT_BROW_POINTS + RIGHT_BROW_POINTS,
NOSE_POINTS + MOUTH_POINTS,
]
FEATHER_AMOUNT = 11
im = np.zeros(sz, dtype=np.float64)
#双眼的外接多边形、鼻和嘴的多边形，作为掩膜
for group in OVERLAY_POINTS:
draw_convex_hull(im,
landmarks[group],
color=1)
# 三维掩码
# im = np.array([im, im, im]).transpose((1, 2, 0))
im = (cv2.GaussianBlur(im, (FEATHER_AMOUNT, FEATHER_AMOUNT), 0) > 0) * 1.0
im = cv2.GaussianBlur(im, (FEATHER_AMOUNT, FEATHER_AMOUNT), 0)
return im.astype('uint8')
def transformation_from_points(target_points, source_points):
# 计算两个点集之间的仿射变换矩阵
# 输入：points1, points2
target_points = target_points.copy()
source_points = source_points.copy()
target_points = target_points.astype(np.float64)
source_points = source_points.astype(np.float64)
c1 = np.mean(target_points, axis=0)
c2 = np.mean(source_points, axis=0)
target_points -= c1
source_points -= c2
s1 = np.std(target_points)
s2 = np.std(source_points)
target_points /= s1
source_points /= s2
U, S, Vt = np.linalg.svd(np.dot(target_points.T , source_points))
R = (np.dot(U , Vt)).T
return np.vstack([np.hstack(((s2 / s1) * R,
np.array([c2.T - np.dot((s2 / s1) * R , c1.T)]).T )),
np.array([0., 0., 1.])])
def wrap_im(im,M,dshape):
'''
对齐图像到目标图上，返回对齐后的targe
'''
output_im = np.zeros(dshape,dtype=im.dtype)
cv2.warpAffine(im,M[:2],(dshape[1],dshape[0]),dst=output_im,borderMode=cv2.BORDER_TRANSPARENT,flags=cv2.WARP_INVERSE_MAP)
return output_im
def correct_colors(im1, im2, landmarks1,COLOUR_CORRECT_BLUR_FRAC = 0.6):
LEFT_EYE_POINTS = list(range(42, 48))
RIGHT_EYE_POINTS = list(range(36, 42))
blur_amount = COLOUR_CORRECT_BLUR_FRAC * np.linalg.norm(
np.mean(landmarks1[LEFT_EYE_POINTS], axis=0) -
np.mean(landmarks1[RIGHT_EYE_POINTS], axis=0))
blur_amount = int(blur_amount)
if blur_amount % 2 == 0:
blur_amount += 1
im1_blur = cv2.GaussianBlur(im1, (blur_amount, blur_amount), 0,0)
im2_blur = cv2.GaussianBlur(im2, (blur_amount, blur_amount), 0,0)
# Avoid divide-by-zero errors.
im2_blur += (128 * (im2_blur <= 1.0)).astype(im2_blur.dtype)
return (im2.astype(np.float64) * im1_blur.astype(np.float64) /im2_blur.astype(np.float64))
def change_face(color_fac):
# 颜色矫正
source_image_align_correct = correct_colors(target_image,source_image_align,tgt_landmarks_np,color_fac)
# 换脸
output_im = (np.multiply(target_image,(1-combined_mask))
+ np.multiply(source_image_align_correct,combined_mask)).astype('uint8')
cv2.imshow('output_im_correct'+str(color_fac), source_image_align_correct.astype('uint8'))
cv2.imshow('output_im_change'+str(color_fac), output_im.astype('uint8'))
if __name__ == '__main__':
plt.ion()
plt.figure('换脸')
# 预训练的68点人脸标记点检测器
predictor_path = os.path.join(pwd_path, "shape_predictor_68_face_landmarks.dat")
predictor = dlib.shape_predictor(predictor_path)
# 读取原始图片和要替换的脸部图片
source_image = cv2.imread(source_img_path)
target_image = cv2.imread(target_img_path)
# 检测人脸特征点
src_landmarks = detect_face_landmarks(source_image, predictor)
tgt_landmarks = detect_face_landmarks(target_image, predictor)
# 显示图片特征点
src_copy = source_image.copy()
src_copy = show_img_landmarks(src_copy, src_landmarks)
tgt_copy = target_image.copy()
tgt_copy = show_img_landmarks(tgt_copy, tgt_landmarks)
# cv2.imshow('source_image', src_copy)
# cv2.imshow('target_image', tgt_copy)
plt.subplot(3,4,11)
plt.title("脸来源图-关键点")
# bgr转换rgb
plt.imshow(src_copy[...,::-1])
plt.subplot(3,4,12)
plt.title("脸目标图-关键点")
plt.imshow(tgt_copy[...,::-1])
# 转换为numpy数组
src_landmarks_np = np.array([[p.x, p.y] for p in src_landmarks.parts()])
tgt_landmarks_np = np.array([[p.x, p.y] for p in tgt_landmarks.parts()])
src_size = (source_image.shape[0], source_image.shape[1])
src_mask = get_face_mask(src_size, src_landmarks_np) # 脸图人脸掩模
plt.subplot(3,4,1)
plt.title("脸来源图")
plt.imshow(source_image[...,::-1])
plt.subplot(3,4,2)
plt.title("脸来源图-掩码")
plt.imshow(src_mask,"gray")
tgt_size = (target_image.shape[0], target_image.shape[1])
tgt_mask = get_face_mask(tgt_size, tgt_landmarks_np) # 脸图人脸掩模
plt.subplot(3,4,3)
plt.title("脸目标图")
plt.imshow(target_image[...,::-1])
plt.subplot(3,4,4)
plt.title("脸目标图-掩码")
plt.imshow(tgt_mask,"gray")
# 计算到目标图的仿射变换矩阵
trans_mat = transformation_from_points(tgt_landmarks_np,src_landmarks_np)
# 仿射变换，对齐来源图像到目标图上
source_image_align = wrap_im(source_image,trans_mat,target_image.shape[:3])
plt.subplot(3,4,5)
plt.title("脸来源图-对齐")
plt.imshow(source_image_align[...,::-1])
# 仿射变换，对齐来源图像掩码到目标图上
src_mask_align = wrap_im(src_mask,trans_mat,target_image.shape[:2])
# 融合掩码
combined_mask = np.max([tgt_mask, src_mask_align],
axis=0)
# combined_mask = np.min([tgt_mask, src_mask_align],
# axis=0)
# combined_mask = tgt_mask
plt.subplot(3,4,6)
plt.title("对齐融合掩码")
combined_mask = np.stack((combined_mask,combined_mask,combined_mask),axis=2)
plt.imshow(combined_mask[...,0],"gray")
plt.subplot(3,4,7)
# combined_mask = combined_mask/255
plt.title("来源图掩码对应底图")
plt.imshow(np.multiply(source_image_align,combined_mask).astype('uint8')[...,::-1])
plt.subplot(3,4,8)
# combined_mask = combined_mask/255
plt.title("目标图掩码对应底图")
plt.imshow(np.multiply(target_image,combined_mask).astype('uint8')[...,::-1])
# 颜色矫正
source_image_align_correct = correct_colors(target_image,source_image_align,tgt_landmarks_np,0.8)
plt.subplot(3,4,9)
plt.title("颜色矫正")
plt.imshow(source_image_align_correct[...,::-1].astype('uint8'))
# 换脸
output_im = (np.multiply(target_image,(1-combined_mask))
+ np.multiply(source_image_align_correct,combined_mask)*1
+ np.multiply(target_image,combined_mask)*0.0 ).astype('uint8')
plt.subplot(3,4,10)
plt.title("换脸")
plt.imshow(cv2.cvtColor(output_im.copy().astype('uint8'),cv2.COLOR_BGR2RGB))
plt.axis('off')
# cv2.imshow('output_im', source_image_align_correct.astype('uint8'))
# cv2.imshow('output_im_correct', output_im.astype('uint8'))
# change_face(0.1)
# change_face(0.5)
# change_face(0.9)
plt.ioff()
plt.show()
# cv2.waitKey(0)
# cv2.destroyAllWindows()
Have a good day!