Python: faces Swap

 

 

?

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231

# encoding: utf-8 # 版权所有 2024 涂聚文有限公司 # 许可信息查看：pip install boost # 描述：pip install boost # pip install dlib # pip install cmake==3.25.2 # pip install dlib==19.24.2  如果安装不上，按此法 # Author    : geovindu,Geovin Du 涂聚文. # IDE       : PyCharm 2023.1 python 3.11 # Datetime  : 2024/6/13 22:09 # User      : geovindu # Product   : PyCharm # Project   : pyBaiduAi # File      : FaceSwaFun.py # explain   : 学习 import cv2 import numpy as np import dlib from PIL import Image as im     class FaceSwaFun(object):     """     换脸类     """       def __init__(self, SOURCEPATH, DESTPATH):         """         实例化         :param SOURCEPATH: 需要用脸的图片         :param DESTPATH: 用脸目标图片         """         self.SOURCE_PATH = SOURCEPATH         self.DEST_PATH = DESTPATH       def index_from_array(self, numpyarray):         """           :param numpyarray:         :return:         """         index = None         for n in numpyarray[0]:             index = n             break         return index       def getImage(self) -> tuple:         """           :return: 返回 (图片的数组，保存的文件名)         """         frontal_face_detector = dlib.get_frontal_face_detector()         frontal_face_predictor = dlib.shape_predictor("dataset/shape_predictor_68_face_landmarks.dat")           source_image = cv2.imread(self.SOURCE_PATH)         source_image_grayscale = cv2.cvtColor(source_image, cv2.COLOR_BGR2GRAY)         #         destination_image = cv2.imread(self.DEST_PATH)         destination_image_grayscale = cv2.cvtColor(destination_image, cv2.COLOR_BGR2GRAY)           source_image_canvas = np.zeros_like(source_image_grayscale)         height, width, no_of_channels = destination_image.shape         destination_image_canvas = np.zeros((height, width, no_of_channels), np.uint8)           source_faces = frontal_face_detector(source_image_grayscale)           # Obtaining source face landmark points, convex hull, creating mask and also getting delaunay triangle face landmark indices for every face         for source_face in source_faces:             source_face_landmarks = frontal_face_predictor(source_image_grayscale, source_face)             source_face_landmark_points = []             for landmark_no in range(68):                 x_point = source_face_landmarks.part(landmark_no).x                 y_point = source_face_landmarks.part(landmark_no).y                 source_face_landmark_points.append((x_point, y_point))               source_face_landmark_points_array = np.array(source_face_landmark_points, np.int32)             source_face_convexhull = cv2.convexHull(source_face_landmark_points_array)               cv2.fillConvexPoly(source_image_canvas, source_face_convexhull, 255)             source_face_image = cv2.bitwise_and(source_image, source_image, mask=source_image_canvas)               # DELAUNAY TRIANGULATION               bounding_rectangle = cv2.boundingRect(source_face_convexhull)             subdivisions = cv2.Subdiv2D(bounding_rectangle)             subdivisions.insert(source_face_landmark_points)             triangles_vector = subdivisions.getTriangleList()             triangles_array = np.array(triangles_vector, dtype=np.int32)               triangle_landmark_points_list = []             source_face_image_copy = source_face_image.copy()               for triangle in triangles_array:                 index_point_1 = (triangle[0], triangle[1])                 index_point_2 = (triangle[2], triangle[3])                 index_point_3 = (triangle[4], triangle[5])                   index_1 = np.where((source_face_landmark_points_array == index_point_1).all(axis=1))                 index_1 = self.index_from_array(index_1)                 index_2 = np.where((source_face_landmark_points_array == index_point_2).all(axis=1))                 index_2 = self.index_from_array(index_2)                 index_3 = np.where((source_face_landmark_points_array == index_point_3).all(axis=1))                 index_3 = self.index_from_array(index_3)                   triangle = [index_1, index_2, index_3]                 triangle_landmark_points_list.append(triangle)           destination_faces = frontal_face_detector(destination_image_grayscale)           # Obtaining destination face landmark points and also convex hull for every face         for destination_face in destination_faces:             destination_face_landmarks = frontal_face_predictor(destination_image_grayscale, destination_face)             destination_face_landmark_points = []             for landmark_no in range(68):                 x_point = destination_face_landmarks.part(landmark_no).x                 y_point = destination_face_landmarks.part(landmark_no).y                 destination_face_landmark_points.append((x_point, y_point))               destination_face_landmark_points_array = np.array(destination_face_landmark_points, np.int32)             destination_face_convexhull = cv2.convexHull(destination_face_landmark_points_array)           # Iterating through all source delaunay triangle and superimposing source triangles in empty destination canvas after warping to same size as destination triangles' shape         for i, triangle_index_points in enumerate(triangle_landmark_points_list):             # Cropping source triangle's bounding rectangle               source_triangle_point_1 = source_face_landmark_points[triangle_index_points[0]]             source_triangle_point_2 = source_face_landmark_points[triangle_index_points[1]]             source_triangle_point_3 = source_face_landmark_points[triangle_index_points[2]]             source_triangle = np.array([source_triangle_point_1, source_triangle_point_2, source_triangle_point_3],                                        np.int32)               source_rectangle = cv2.boundingRect(source_triangle)             (x, y, w, h) = source_rectangle             cropped_source_rectangle = source_image[y:y + h, x:x + w]               source_triangle_points = np.array([[source_triangle_point_1[0] - x, source_triangle_point_1[1] - y],                                                [source_triangle_point_2[0] - x, source_triangle_point_2[1] - y],                                                [source_triangle_point_3[0] - x, source_triangle_point_3[1] - y]],                                               np.int32)               # Create a mask using cropped destination triangle's bounding rectangle(for same landmark points as used for source triangle)               destination_triangle_point_1 = destination_face_landmark_points[triangle_index_points[0]]             destination_triangle_point_2 = destination_face_landmark_points[triangle_index_points[1]]             destination_triangle_point_3 = destination_face_landmark_points[triangle_index_points[2]]             destination_triangle = np.array(                 [destination_triangle_point_1, destination_triangle_point_2, destination_triangle_point_3], np.int32)               destination_rectangle = cv2.boundingRect(destination_triangle)             (x, y, w, h) = destination_rectangle               cropped_destination_rectangle_mask = np.zeros((h, w), np.uint8)               destination_triangle_points = np.array(                 [[destination_triangle_point_1[0] - x, destination_triangle_point_1[1] - y],                  [destination_triangle_point_2[0] - x, destination_triangle_point_2[1] - y],                  [destination_triangle_point_3[0] - x, destination_triangle_point_3[1] - y]],                 np.int32)               cv2.fillConvexPoly(cropped_destination_rectangle_mask, destination_triangle_points, 255)               # Warp source triangle to match shape of destination triangle and put it over destination triangle mask               source_triangle_points = np.float32(source_triangle_points)             destination_triangle_points = np.float32(destination_triangle_points)               matrix = cv2.getAffineTransform(source_triangle_points, destination_triangle_points)             warped_rectangle = cv2.warpAffine(cropped_source_rectangle, matrix, (w, h))               warped_triangle = cv2.bitwise_and(warped_rectangle, warped_rectangle,                                               mask=cropped_destination_rectangle_mask)               # Reconstructing destination face in empty canvas of destination image               # removing white lines in triangle using masking             new_dest_face_canvas_area = destination_image_canvas[y:y + h, x:x + w]             new_dest_face_canvas_area_gray = cv2.cvtColor(new_dest_face_canvas_area, cv2.COLOR_BGR2GRAY)             _, mask_created_triangle = cv2.threshold(new_dest_face_canvas_area_gray, 1, 255, cv2.THRESH_BINARY_INV)               warped_triangle = cv2.bitwise_and(warped_triangle, warped_triangle, mask=mask_created_triangle)             new_dest_face_canvas_area = cv2.add(new_dest_face_canvas_area, warped_triangle)             destination_image_canvas[y:y + h, x:x + w] = new_dest_face_canvas_area           # Put reconstructed face on the destination image         final_destination_canvas = np.zeros_like(destination_image_grayscale)         final_destination_face_mask = cv2.fillConvexPoly(final_destination_canvas, destination_face_convexhull, 255)         final_destination_canvas = cv2.bitwise_not(final_destination_face_mask)         destination_face_masked = cv2.bitwise_and(destination_image, destination_image, mask=final_destination_canvas)         destination_with_face = cv2.add(destination_face_masked, destination_image_canvas)           # Seamless cloning to make attachment blend with surrounding pixels           # we have to find center point of reconstructed convex hull to pass into seamlessClone()         (x, y, w, h) = cv2.boundingRect(destination_face_convexhull)         destination_face_center_point = (int((x + x + w) / 2), int((y + y + h) / 2))         seamless_cloned_face = cv2.seamlessClone(destination_with_face, destination_image, final_destination_face_mask,                                                  destination_face_center_point, cv2.NORMAL_CLONE)           data = im.fromarray(seamless_cloned_face)           # saving the final output         # as a PNG file         file = 'geovindu.png'         data.save(file)           # cv2.imshow("Destination image with source face 2", seamless_cloned_face)         # cv2.waitKey(0)         # cv2.destroyAllWindows()         print(type(seamless_cloned_face))         return (seamless_cloned_face, file)           '''         1.         import cv2         cv2.imwrite("geovindu.jpg", seamless_cloned_face)         2.         from PIL import Image         im = Image.fromarray(seamless_cloned_face)         im.save("geovindu.jpg")         3.         import scipy.misc         scipy.misc.imsave('geovindu.jpg', seamless_cloned_face)         4.         import scipy.misc         scipy.misc.toimage(seamless_cloned_face, cmin=0.0, cmax=...).save('geovindu.jpg')         5.         import matplotlib         matplotlib.image.imsave('geovindu.png', seamless_cloned_face)           '''

　　

 

 

调用：

?

1 2 3 4 5 6 7

# 调用     du= BLL.imageFaceSwapFun.FaceSwaFun("media/images/modi.jpg","media/images/viplav.jpeg")     #du.SOURCE_PATH="media/images/modi.jpg"     # du.DEST_PATH="media/images/viplav.jpeg"     geovindu=du.getImage()     print(geovindu)     cv2.imwrite("geovindu20.png", geovindu[0])