学习GitHub上比较火换脸博客，原英文版：https://matthewearl.github.io/2015/07/28/switching-eds-with-python/
系统win10，x64
- 安装python 2.7
- opencv3.0下载，安装，配置环境变量（所需python版本为2.7）
- 下载numpy，版本numpy-1.10.2-win32-superpack-python2.7，必须与python版本一致，不然即使找到了cv模块也不能够运行。
- opencv文件夹中，build->python->2.7 复制2.7下面的所有文件 到C:Python27Libsite-packages 中
- 测试是否配置成功：
import cv2 image = cv2.imread("0.png") cv2.imshow("Image",image) cv2.waitKey(0)
开始学习换脸：
- 下载boost，编译boost：解压，执行bootstrap.bat（使用vs2015编译），会在boost根目录生成 b2.exe 、bjam.exe 、project-config.jam 、bootstrap.log四个文件，其中，b2.exe 、bjam.exe 这两个exe作用是一样的，bjam.exe 是老版本，b2是bjam的升级版本。运行bjam.exe，编译c++版本的boost库，配置环境变量BOOST_ROOT=C:oost_1_60_0；BOOST_LIBRARYDIR=C:oost_1_60_0stagelib。再编译python动态链接库，b2.exe --with-python --build-type=complete。
- 下载dlib从http://dlib.net/，Dlib is a modern C++ toolkit containing machine learning algorithms and tools for creating complex software in C++ to solve real world problems.编译python API，命令python setup.py install
- 使用dlib抽取脸部标志点：Dlib实现了paper ”one millisecond face alignment with an ensemble of regression trees" by Vahid Kazemi and Josephine Sullivan. 虽然算法本身很复杂，但是它的python接口的使用很简单：
1 import cv2 2 import dlib 3 import numpy 4 import sys 5 6 PREDICTOR_PATH = "shape_predictor_68_face_landmarks.dat" 7 SCALE_FACTOR = 1 8 FEATURE_AMOUNT = 11 9 10 FACE_POINTS = list(range(17, 68)) 11 MOUTH_POINTS = list(range(48, 68)) 12 RIGHT_BROW_POINTS = list(range(17, 22)) 13 LEFT_BROW_POINTS = list(range(22, 27)) 14 RIGHT_EYE_POINTS = list(range(36, 42)) 15 LEFT_EYE_POINTS = list(range(42, 48)) 16 NOSE_POINTS = list(range(27, 35)) 17 JAW_POINTS = list(range(0, 17)) 18 19 # Points used to line up the images 20 ALIGN_POINTS = (LEFT_BROW_POINTS + RIGHT_EYE_POINTS + LEFT_EYE_POINTS + 21 RIGHT_BROW_POINTS + NOSE_POINTS + MOUTH_POINTS) 22 23 # Points from the second image to overlay on the first. The convex hull of 24 # each element will be overlaid 25 OVERLAY_POINTS = [ 26 LEFT_EYE_POINTS + RIGHT_EYE_POINTS + LEFT_BROW_POINTS 27 + RIGHT_BROW_POINTS, 28 NOSE_POINTS + MOUSE_POINTS, 29 ] 30 31 # Amount of blur to use during color correction, as a fraction of the 32 # pupillary distance 33 COLOUR_CORRECT_BLUR_FRAC = 0.6 34 35 detector = dlib.get_frontal_face_detector() 36 predictor = dlib.shape_predictor(PREDICTOR_PATH) 37 38 class TooManyFaces(Exception): 39 pass 40 41 class NoFaces(Exception): 42 pass 43 44 ## input: an image in the form of a numpy array 45 ## return: a 68 * 2 element matrix, each row corresponding with 46 ## the x, y coordintes of a pariticular feature point in the input image 47 def get_landmarks(im): 48 rects = detector(im, 1) 49 50 if len(rects) > 1: 51 raise TooManyFaces 52 if len(rects) == 0: 53 raise NoFaces 54 55 # the feature extractor (predictor) requires a rough bounding box as input 56 # to the algorithm. This is provided by a traditional face detector ( 57 # detector) which returns a list of rectangles, each of which corresponding 58 # a face in the image 59 return numpy.matrix([[p.x p.y] for p in predictor(im, rects[0]).parts()])
为了使用predictor，需要利用一个提前训练好的model：shape_predictor_68_face_landmarks.dat，从http://sourceforge.net/projects/dclib/files/dlib/v18.10/shape_predictor_68_face_landmarks.dat.bz2下载
4. 用Procrustes Analysis进行脸部对准：目前我们已经有两个人脸的landmark矩阵，矩阵的每一行代表一个脸部特征的坐标。现在我们要做的是找出如何通过旋转、平移、和尺度操作使得第一张脸的特征点与第二张脸的尽可能的匹配。找到这个合适的匹配变换之后，我们就可以将第二张脸用同样的变换覆盖第一张脸。
从数学上考虑，我们寻找平移参数T，尺度参数s，和旋转变换矩阵R使得如下目标函数
最小化，其中R是2*2的正交矩阵，s是标量，T是2*1的向量，pi和qi是landmark矩阵的行（对应的脸部特征坐标）。
这个问题可以被Ordinary Procrustes Analysis求解。
-
def transformation_from_points(points1, points2): """ Return an affine transformation [s * R | T] such that: sum || s*R*p1,i + T - p2,i||^2 is minimized. """ # Solve the procrustes problem by substracting centroids, scaling by the # standard deviation, and then using the SVD to calculate the rotation. See # the following for more details: # https://en.wikipedia.org/wiki/Orthogonal_Procrustes_problem points1 = points1.astype(numpy.float64) points2 = points2.astype(numpy.float64) c1 = numpy.mean(points1, axis=0) c2 = numpy.mean(points2, axis=0) points1 -= c1 points2 -= c2 s1 = numpy.std(points1) s2 = numpy.std(points2) points1 /= s1 points2 /= s2 U, S, Vt = numpy.linalg.svd(points1.T * points2) # The R we seek is in fact the transpose of the one given by U * Vt. This # is because the above formulation assumes the matrix goes on the right # (with row vectors) where as our solution requires the matrix to be on the # left (with column vectors). R = (U * Vt).T return numpy.vstack([numpy.hstack(((s2 / s1) * R, c2.T - (s2 / s1) * R * c1.T)), numpy.matrix([0., 0., 1.])])
求解步骤：
1) 将输入矩阵转化为浮点型，这一操作被后面步骤需要；
2) 每个点集减去中心点（去中心操作）；
3) 每个点集除以标准差，解决尺度问题；
4) 使用SVD (Singular Value Decomposition) 计算旋转矩阵，解Orthogonal Procrustes Problem;
5) 返回完整的仿射变换矩阵，维度3* 3.
获得的仿射变换可以应用到第二幅图像，与第一张图像匹配：
1 def warp_im(im, M, dshape): 2 output_im = numpy.zeros(dshape, dtype=im.dtype) 3 cv2.warpAffine(im, 4 M[:2], 5 (dshape[1], dshape[0]), 6 dst=output_im, 7 borderMode=cv2.BORDER_TRANSPARENT, 8 flags=cv2.WARP_INVERSE_MAP) 9 return output_im
5. 计算mask，并进行色彩校正：利用眼部和眉毛区域特征点计算二维凸包，鼻子和嘴部特征点再计算二维凸包，获得一个五官的mask，代码和结果如下：
1 def draw_convex_hull(im, points, color): 2 points = cv2.convexHull(points) 3 cv2.fillConvexPoly(im, points, color=color) 4 5 def get_face_mask(im, landmarks): 6 im = numpy.zeros(im.shape[:2], dtype=numpy.float64) 7 8 for group in OVERLAY_POINTS: 9 draw_convex_hull(im, 10 landmarks[group], 11 color=1) 12 13 im = numpy.array([im, im, im]).transpose((1, 2, 0)) 14 15 im = (cv2.GaussianBlur(im, (FEATURE_AMOUNT, FEATURE_AMOUNT), 0) > 0) * 1.0 16 im = cv2.GaussianBlur(im, (FEATURE_AMOUNT, FEATURE_AMOUNT), 0) 17 18 return im
如果我们直接将脸部mask区域覆盖，我们会发现脸部颜色不一致的问题：
进行色彩矫正，改变第二张脸的颜色使其可以与第一张脸匹配。做法是将第二张脸的颜色除以第二张脸的高斯模糊值，再乘以第一张脸的高斯模糊值，点操作。参考https://en.wikipedia.org/wiki/Color_balance#Scaling_monitor_R.2C_G.2C_and_B，并没有将整幅图像乘以常数因子，而是将每个像素乘以它自己的尺度因子。
通过这个操作，可以一定程度上弥补两幅图像之间的亮度不同问题。代码如下：
def correct_colors(im1, im2, landmarks1): blur_amount = COLOUR_CORRECT_BLUR_FRAC * numpy.linalg.norm( numpy.mean(landmarks1[LEFT_EYE_POINTS], axis=0) - numpy.mean(landmarks2[RIGHT_EYE_POINTS], axis=0)) blur_amount = int(blur_amount) if blur_amount % 2 == 0: blur_amount += 1 print blur_amount im1_blur = cv2.GaussianBlur(im1, (blur_amount, blur_amount), 0) im2_blur = cv2.GaussianBlur(im2, (blur_amount, blur_amount), 0) cv2.imshow("Image", im1_blur) # warp_im(im2, M, im1.shape) cv2.waitKey(0) cv2.imshow("Image", im2_blur) # warp_im(im2, M, im1.shape) cv2.waitKey(0) # Avoid divide-by-zero errors: im2_blur += (128 * (im2_blur <= 1.0)).astype(im2_blur.dtype) cv2.imshow("Image", im2_blur) # warp_im(im2, M, im1.shape) cv2.waitKey(0) cv2.destroyWindow("Image") return (im2.astype(numpy.float64) * im1_blur.astype(numpy.float64) / im2_blur.astype(numpy.float64))
这种做法可以在粗略地解决色彩不一致问题，效果与高斯kernel的大小密切相关：kernel太小，第一张脸中本应该被覆盖的脸部特征会出现在最后的融合图中；kernel太大，第二张脸外部的像素会被引入融合图像，产生污点。下图的kernel size等于0.05*瞳间距。
6. 融合：将经过色彩矫正的第二张脸的mask区域与第一张脸融合：
output_im = im1 * (1.0 - combined_mask) + warped_corrected_im2 * combined_mask
至此换脸全部完成，全部代码如下：
import cv2 import dlib import numpy import sys PREDICTOR_PATH = "shape_predictor_68_face_landmarks.dat" SCALE_FACTOR = 1 FEATURE_AMOUNT = 11 FACE_POINTS = list(range(17, 68)) MOUTH_POINTS = list(range(48, 61)) RIGHT_BROW_POINTS = list(range(17, 22)) LEFT_BROW_POINTS = list(range(22, 27)) RIGHT_EYE_POINTS = list(range(36, 42)) LEFT_EYE_POINTS = list(range(42, 48)) NOSE_POINTS = list(range(27, 35)) JAW_POINTS = list(range(0, 17)) # Points used to line up the images ALIGN_POINTS = (LEFT_BROW_POINTS + RIGHT_EYE_POINTS + LEFT_EYE_POINTS + RIGHT_BROW_POINTS + NOSE_POINTS + MOUTH_POINTS) # Points from the second image to overlay on the first. The convex hull of # each element will be overlaid OVERLAY_POINTS = [ LEFT_EYE_POINTS + RIGHT_EYE_POINTS + LEFT_BROW_POINTS + RIGHT_BROW_POINTS, NOSE_POINTS + MOUTH_POINTS, ] # Amount of blur to use during color correction, as a fraction of the # pupillary distance COLOUR_CORRECT_BLUR_FRAC = 0.05 detector = dlib.get_frontal_face_detector() predictor = dlib.shape_predictor(PREDICTOR_PATH) class TooManyFaces(Exception): pass class NoFaces(Exception): pass ## input: an image in the form of a numpy array ## return: a 68 * 2 element matrix, each row corresponding with ## the x, y coordintes of a pariticular feature point in the input image def get_landmarks(im): rects = detector(im, 1) if len(rects) > 1: raise TooManyFaces if len(rects) == 0: raise NoFaces # the feature extractor (predictor) requires a rough bounding box as input # to the algorithm. This is provided by a traditional face detector ( # detector) which returns a list of rectangles, each of which corresponding # a face in the image return numpy.matrix([[p.x, p.y] for p in predictor(im, rects[0]).parts()]) def annote_landmarks(im, landmarks): im = im.copy() for idx, point in enumerate(landmarks): pos = (point[0, 0], point[0, 1]) cv2.putText(im, str(idx), pos, fontFace=cv2.FONT_HERSHEY_SCRIPT_SIMPLEX, fontScale=0.4, color=(0, 0, 255)) cv2.circle(im, pos, 3, color=(0, 255, 255)) return im def read_im_and_landmarks(fname): im = cv2.imread(fname, cv2.IMREAD_COLOR) im = cv2.resize(im, (im.shape[1] * SCALE_FACTOR, im.shape[0] * SCALE_FACTOR)) s = get_landmarks(im) return im, s def transformation_from_points(points1, points2): """ Return an affine transformation [s * R | T] such that: sum || s*R*p1,i + T - p2,i||^2 is minimized. """ # Solve the procrustes problem by substracting centroids, scaling by the # standard deviation, and then using the SVD to calculate the rotation. See # the following for more details: # https://en.wikipedia.org/wiki/Orthogonal_Procrustes_problem points1 = points1.astype(numpy.float64) points2 = points2.astype(numpy.float64) c1 = numpy.mean(points1, axis=0) c2 = numpy.mean(points2, axis=0) points1 -= c1 points2 -= c2 s1 = numpy.std(points1) s2 = numpy.std(points2) points1 /= s1 points2 /= s2 U, S, Vt = numpy.linalg.svd(points1.T * points2) # The R we seek is in fact the transpose of the one given by U * Vt. This # is because the above formulation assumes the matrix goes on the right # (with row vectors) where as our solution requires the matrix to be on the # left (with column vectors). R = (U * Vt).T return numpy.vstack([numpy.hstack(((s2 / s1) * R, c2.T - (s2 / s1) * R * c1.T)), numpy.matrix([0., 0., 1.])]) def draw_convex_hull(im, points, color): points = cv2.convexHull(points) cv2.fillConvexPoly(im, points, color=color) def get_face_mask(im, landmarks): im = numpy.zeros(im.shape[:2], dtype=numpy.float64) for group in OVERLAY_POINTS: draw_convex_hull(im, landmarks[group], color=1) im = numpy.array([im, im, im]).transpose((1, 2, 0)) im = (cv2.GaussianBlur(im, (FEATURE_AMOUNT, FEATURE_AMOUNT), 0) > 0) * 1.0 im = cv2.GaussianBlur(im, (FEATURE_AMOUNT, FEATURE_AMOUNT), 0) return im def warp_im(im, M, dshape): output_im = numpy.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): blur_amount = COLOUR_CORRECT_BLUR_FRAC * numpy.linalg.norm( numpy.mean(landmarks1[LEFT_EYE_POINTS], axis=0) - numpy.mean(landmarks2[RIGHT_EYE_POINTS], axis=0)) blur_amount = int(blur_amount) if blur_amount % 2 == 0: blur_amount += 1 print blur_amount im1_blur = cv2.GaussianBlur(im1, (blur_amount, blur_amount), 0) im2_blur = cv2.GaussianBlur(im2, (blur_amount, blur_amount), 0) # Avoid divide-by-zero errors: im2_blur += (128 * (im2_blur <= 1.0)).astype(im2_blur.dtype) return (im2.astype(numpy.float64) * im1_blur.astype(numpy.float64) / im2_blur.astype(numpy.float64)) im1, landmarks1 = read_im_and_landmarks("0.jpg") im2, landmarks2 = read_im_and_landmarks("1.jpg") # draw landmarks ##for i in landmarks2: ## im2[i[0,1], i[0,0]] = [0,0,0] ##cv2.imshow("Image0", annote_landmarks(im1, landmarks1)) ##cv2.waitKey(0) ##cv2.destroyWindow("Image0") ##cv2.imshow("Image1", annote_landmarks(im2, landmarks2)) ##cv2.waitKey(0) M = transformation_from_points(landmarks1[ALIGN_POINTS], landmarks2[ALIGN_POINTS]) mask = get_face_mask(im2, landmarks2) warped_mask = warp_im(mask, M, im1.shape) combined_mask = numpy.max([get_face_mask(im1, landmarks1), warped_mask], axis=0) warped_im2 = warp_im(im2, M, im1.shape) warped_corrected_im2 = correct_colors(im1, warped_im2, landmarks1) output_im = im1 * (1.0 - combined_mask) + warped_corrected_im2 * combined_mask cv2.imshow("Image1", output_im.astype(output_im.dtype)) # warp_im(im2, M, im1.shape) cv2.waitKey(0) cv2.destroyWindow("Image1") cv2.imwrite("output.jpg", output_im)