在這篇文章中我將介紹如何寫一個(gè)簡(jiǎn)短(200行)的 Python 腳本,來(lái)自動(dòng)地將一幅圖片的臉替換為另一幅圖片的臉。
這個(gè)過(guò)程分四步:
檢測(cè)臉部標(biāo)記。
旋轉(zhuǎn)、縮放、平移和第二張圖片,以配合第一步。
調(diào)整第二張圖片的色彩平衡,以適配第一張圖片。
把第二張圖像的特性混合在第一張圖像中。
1.使用dlib提取面部標(biāo)記
該腳本使用dlib的 Python 綁定來(lái)提取面部標(biāo)記:
Dlib 實(shí)現(xiàn)了 Vahid Kazemi 和 Josephine Sullivan 的《使用回歸樹一毫秒臉部對(duì)準(zhǔn)》論文中的算法。算法本身非常復(fù)雜,但dlib接口使用起來(lái)非常簡(jiǎn)單:
PREDICTOR_PATH = "/home/matt/dlib-18.16/shape_predictor_68_face_landmarks.dat"
detector = dlib.get_frontal_face_detector()
predictor = dlib.shape_predictor(PREDICTOR_PATH)
defget_landmarks(im):
rects = detector(im,1)
iflen(rects) >1:
raiseTooManyFaces
iflen(rects) == 0:
raiseNoFaces
returnnumpy.matrix([[p.x,p.y]forpinpredictor(im,rects[0]).parts()])
get_landmarks()函數(shù)將一個(gè)圖像轉(zhuǎn)化成numpy數(shù)組,并返回一個(gè)68×2元素矩陣,輸入圖像的每個(gè)特征點(diǎn)對(duì)應(yīng)每行的一個(gè)x,y坐標(biāo)。
特征提取器(predictor)需要一個(gè)粗糙的邊界框作為算法輸入,由一個(gè)傳統(tǒng)的能返回一個(gè)矩形列表的人臉檢測(cè)器(detector)提供,其每個(gè)矩形列表在圖像中對(duì)應(yīng)一個(gè)臉。
2.用 Procrustes 分析調(diào)整臉部
現(xiàn)在我們已經(jīng)有了兩個(gè)標(biāo)記矩陣,每行有一組坐標(biāo)對(duì)應(yīng)一個(gè)特定的面部特征(如第30行的坐標(biāo)對(duì)應(yīng)于鼻頭)。我們現(xiàn)在要解決如何旋轉(zhuǎn)、翻譯和縮放第一個(gè)向量,使它們盡可能適配第二個(gè)向量的點(diǎn)。一個(gè)想法是可以用相同的變換在第一個(gè)圖像上覆蓋第二個(gè)圖像。
將這個(gè)問題數(shù)學(xué)化,尋找T,s和R,使得下面這個(gè)表達(dá)式:
結(jié)果最小,其中R是個(gè)2×2正交矩陣,s是標(biāo)量,T是二維向量,pi和qi是上面標(biāo)記矩陣的行。
事實(shí)證明,這類問題可以用“常規(guī) Procrustes 分析法”解決:
deftransformation_from_points(points1,points2):
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)
R = (U * Vt).T
returnnumpy.vstack([numpy.hstack(((s2 / s1) * R,
c2.T - (s2 / s1) * R * c1.T)),
numpy.matrix([0.,0.,1.])])
代碼實(shí)現(xiàn)了這幾步:
1.將輸入矩陣轉(zhuǎn)換為浮點(diǎn)數(shù)。這是后續(xù)操作的基礎(chǔ)。
2.每一個(gè)點(diǎn)集減去它的矩心。一旦為點(diǎn)集找到了一個(gè)最佳的縮放和旋轉(zhuǎn)方法,這兩個(gè)矩心c1和c2就可以用來(lái)找到完整的解決方案。
3.同樣,每一個(gè)點(diǎn)集除以它的標(biāo)準(zhǔn)偏差。這會(huì)消除組件縮放偏差的問題。
4.使用奇異值分解計(jì)算旋轉(zhuǎn)部分。可以在維基百科上看到關(guān)于解決正交 Procrustes 問題的細(xì)節(jié)。
5.利用仿射變換矩陣返回完整的轉(zhuǎn)化。
其結(jié)果可以插入 OpenCV 的cv2.warpAffine函數(shù),將圖像二映射到圖像一:
defwarp_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)
returnoutput_im
對(duì)齊結(jié)果如下:
3.校正第二張圖像的顏色
如果我們?cè)噲D直接覆蓋面部特征,很快會(huì)看到這個(gè)問題:
這個(gè)問題是兩幅圖像之間不同的膚色和光線造成了覆蓋區(qū)域的邊緣不連續(xù)。我們?cè)囍拚?/p>
COLOUR_CORRECT_BLUR_FRAC = 0.6
LEFT_EYE_POINTS = list(range(42,48))
RIGHT_EYE_POINTS = list(range(36,42))
defcorrect_colours(im1,im2,landmarks1):
blur_amount = COLOUR_CORRECT_BLUR_FRAC * numpy.linalg.norm(
numpy.mean(landmarks1[LEFT_EYE_POINTS],axis=0) -
numpy.mean(landmarks1[RIGHT_EYE_POINTS],axis=0))
blur_amount = int(blur_amount)
ifblur_amount % 2 == 0:
blur_amount += 1
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)
return(im2.astype(numpy.float64) * im1_blur.astype(numpy.float64) /
im2_blur.astype(numpy.float64))
結(jié)果如下:
此函數(shù)試圖改變 im2 的顏色來(lái)適配 im1。它通過(guò)用 im2 除以 im2 的高斯模糊值,然后乘以im1的高斯模糊值。這里的想法是用RGB縮放校色,但并不是用所有圖像的整體常數(shù)比例因子,每個(gè)像素都有自己的局部比例因子。
用這種方法兩圖像之間光線的差異只能在某種程度上被修正。例如,如果圖像1是從一側(cè)照亮,但圖像2是被均勻照亮的,色彩校正后圖像2也會(huì)出現(xiàn)未照亮一側(cè)暗一些的問題。
也就是說(shuō),這是一個(gè)相當(dāng)簡(jiǎn)陋的辦法,而且解決問題的關(guān)鍵是一個(gè)適當(dāng)?shù)母咚购撕瘮?shù)大小。如果太小,第一個(gè)圖像的面部特征將顯示在第二個(gè)圖像中。過(guò)大,內(nèi)核之外區(qū)域像素被覆蓋,并發(fā)生變色。這里的內(nèi)核用了一個(gè)0.6 *的瞳孔距離。
4.把第二張圖像的特征混合在第一張圖像中
用一個(gè)遮罩來(lái)選擇圖像2和圖像1的哪些部分應(yīng)該是最終顯示的圖像:
值為1(顯示為白色)的地方為圖像2應(yīng)該顯示出的區(qū)域,值為0(顯示為黑色)的地方為圖像1應(yīng)該顯示出的區(qū)域。值在0和1之間為圖像1和圖像2的混合區(qū)域。
這是生成上圖的代碼:
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
defdraw_convex_hull(im,points,color):
points = cv2.convexHull(points)
cv2.fillConvexPoly(im,points,color=color)
defget_face_mask(im,landmarks):
im = numpy.zeros(im.shape[:2],dtype=numpy.float64)
forgroup inOVERLAY_POINTS:
draw_convex_hull(im,
landmarks[group],
color=1)
im = numpy.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)
returnim
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)
我們把上述過(guò)程分解:
get_face_mask()的定義是為一張圖像和一個(gè)標(biāo)記矩陣生成一個(gè)遮罩,它畫出了兩個(gè)白色的凸多邊形:一個(gè)是眼睛周圍的區(qū)域,一個(gè)是鼻子和嘴部周圍的區(qū)域。之后它由11個(gè)像素向遮罩的邊緣外部羽化擴(kuò)展,可以幫助隱藏任何不連續(xù)的區(qū)域。
這樣一個(gè)遮罩同時(shí)為這兩個(gè)圖像生成,使用與步驟2中相同的轉(zhuǎn)換,可以使圖像2的遮罩轉(zhuǎn)化為圖像1的坐標(biāo)空間。
之后,通過(guò)一個(gè)element-wise最大值,這兩個(gè)遮罩結(jié)合成一個(gè)。結(jié)合這兩個(gè)遮罩是為了確保圖像1被掩蓋,而顯現(xiàn)出圖像2的特性。
最后,使用遮罩得到最終的圖像:
output_im = im1 * (1.0 - combined_mask) + warped_corrected_im2 * combined_mask
完整代碼(link):
importcv2
importdlib
importnumpy
importsys
PREDICTOR_PATH = "/home/matt/dlib-18.16/shape_predictor_68_face_landmarks.dat"
SCALE_FACTOR = 1
FEATHER_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 colour correction, as a fraction of the
# pupillary distance.
COLOUR_CORRECT_BLUR_FRAC = 0.6
detector = dlib.get_frontal_face_detector()
predictor = dlib.shape_predictor(PREDICTOR_PATH)
classTooManyFaces(Exception):
pass
classNoFaces(Exception):
pass
defget_landmarks(im):
rects = detector(im,1)
iflen(rects) > 1:
raiseTooManyFaces
iflen(rects) == 0:
raiseNoFaces
returnnumpy.matrix([[p.x,p.y]forpinpredictor(im,rects[0]).parts()])
defannotate_landmarks(im,landmarks):
im = im.copy()
foridx,point inenumerate(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))
returnim
defdraw_convex_hull(im,points,color):
points = cv2.convexHull(points)
cv2.fillConvexPoly(im,points,color=color)
defget_face_mask(im,landmarks):
im = numpy.zeros(im.shape[:2],dtype=numpy.float64)
forgroup inOVERLAY_POINTS:
draw_convex_hull(im,
landmarks[group],
color=1)
im = numpy.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)
returnim
deftransformation_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 subtracting 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
returnnumpy.vstack([numpy.hstack(((s2 / s1) * R,
c2.T - (s2 / s1) * R * c1.T)),
numpy.matrix([0.,0.,1.])])
defread_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)
returnim,s
defwarp_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)
returnoutput_im
defcorrect_colours(im1,im2,landmarks1):
blur_amount = COLOUR_CORRECT_BLUR_FRAC * numpy.linalg.norm(
numpy.mean(landmarks1[LEFT_EYE_POINTS],axis=0) -
numpy.mean(landmarks1[RIGHT_EYE_POINTS],axis=0))
blur_amount = int(blur_amount)
ifblur_amount % 2 == 0:
blur_amount += 1
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)
return(im2.astype(numpy.float64) * im1_blur.astype(numpy.float64) /
im2_blur.astype(numpy.float64))
im1,landmarks1 = read_im_and_landmarks(sys.argv[1])
im2,landmarks2 = read_im_and_landmarks(sys.argv[2])
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_colours(im1,warped_im2,landmarks1)
output_im = im1 * (1.0 - combined_mask) + warped_corrected_im2 * combined_mask
cv2.imwrite('output.jpg',output_im)
-
代碼
+關(guān)注
關(guān)注
30文章
4788瀏覽量
68611 -
python
+關(guān)注
關(guān)注
56文章
4797瀏覽量
84688
原文標(biāo)題:小 200 行 Python 代碼做了一個(gè)換臉程序
文章出處:【微信號(hào):magedu-Linux,微信公眾號(hào):馬哥Linux運(yùn)維】歡迎添加關(guān)注!文章轉(zhuǎn)載請(qǐng)注明出處。
發(fā)布評(píng)論請(qǐng)先 登錄
相關(guān)推薦
評(píng)論