人脸识别有趣应用3——抠出人脸并给人脸涂上戏剧脸谱_人脸识别可以按正方形框抠出人脸图吗

前言

人脸最基础的操作之一，是要将人脸识别出来后，把真个人脸给抠出来，这样就可以对人脸进行各种操作，比如：美白、去痘等等，本篇是基于人脸识别库，结合阈值分割图片的mask操作以及一些两个图片的叠加操作，实现了给人脸叠加上京剧脸谱的效果：
原图如下（由AI生成不涉及肖像权）：

经过处理如下：

原理

本篇处理的流程思路如下：
Step1、首先利用face_recognition库对人脸进行识别，并识别出人物的下部脸轮廓，并利用opencv的多边形绘制工具polylines，绘制出人物脸部的下半部轮廓，由于识别库没有提供上半部轮廓的识别功能，暂时用opencv的矩形绘制工具rectangle，把人脸的上半部以矩形表示，总体形成一个mask：

Step2、那么接下来，就是要将人脸的上半部分轮廓给找出来，这样就可以找到整个脸的轮廓，这里用的方法是通过简单的阈值分割，对人脸原图进行阈值分割，并进行形态学的膨胀操作，这样可以把发际线给简单找出来（如下图标红部分）：

Step3:接下来，利用以上两步找到的mask信息，对准备好的京剧脸谱图像进行运算，结果是把原图中的人脸多余部分给去除掉，剩下匹配人脸的部分：

Step4：接下来，就是跟上几节原理一样，把具有alpha通道的png脸谱图（可以通过ps做出来），通过眼睛和嘴巴的定位，较为准确地放到人脸上面，我这里眼睛比较对位了，但是鼻子还需要优化一下。通过一个规则逻辑进行绘制（线性叠加）：
合成地像素点值=人脸原图像素*（1-alpha值）+alpha值*((1-人脸mask值)脸谱像素+人脸mask值人脸原图像素)
对应源代码中的：
for c in range(0,3):
backimg[y1:y2, x1:x2, c] = (alpha_huzijbackimg[y1:y2,x1:x2,c]) + (alpha_huzip((maskalbackimg[y1:y2,x1:x2,c]) + mask3[y1:y2,x1:x2]((1-ddratio)backimg[y1:y2,x1:x2,c]+ddratio(mask[:usy,:usx,c]))))
结束
PS:后续需要优化，现在看起来不是很自然，可以通过边缘的模糊，纹理的叠加，以及一些变形的操作，把脸谱与人脸做到更加精细的叠加。

Python源代码

# -*- coding: utf-8 -*-
"""
Created on Sat Apr 23 17:13:28 2022

@author: JAMES FEI 
Copyright (C) 2022 FEI PANFENG, All rights reserved.
THIS SOFTEWARE, INCLUDING DOCUMENTATION,IS PROTECTED BY COPYRIGHT CONTROLLED 
BY FEI PANFENG ALL RIGHTS ARE RESERVED.
"""
import face_recognition
import cv2
import numpy as np
import time

video_capture = cv2.VideoCapture(0)

def threshold(inputimg,midthreshold=127,maxthreshold=255,binarymod=cv2.THRESH_BINARY):
  
    img=inputimg    
    if len(img.shape)==3:
        # 将图片转为灰度图
        img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) 
 
    retval, output = cv2.threshold(img,midthreshold, maxthreshold, binarymod)
    #print(output.shape)
    return output

process_this_frame = True
 #载入胡子，带透明通道
maskk = cv2.imread('mask.png', cv2.IMREAD_UNCHANGED)

def putonmask(mask,backimg,mask3,dilated,eyeline,noseline,me=140,ne=126,lefteye=(162,258),leftcenter=(50,50)):  
    
    height=mask.shape[0]
    width=mask.shape[1]
    
    ratiow=eyeline/140
    ratioh=noseline/126
    
    
    print("mask shape",mask.shape,ratiow,ratioh)
    mask = cv2.resize(mask,(int(width*ratiow),int(height*ratioh)),interpolation=cv2.INTER_LINEAR)  
    
    lefteye_mask=(int(lefteye[0]*(eyeline/140)),int(lefteye[1]*(noseline/126)))
    
    (x1,y1)=(leftcenter[0]-lefteye_mask[0],leftcenter[1]-lefteye_mask[1])
    
    x2 = x1 + mask.shape[1]
    y2 = y1 + mask.shape[0] 
    usy=mask.shape[0] 
    usx=mask.shape[1]
    
    if x2>backimg.shape[1]:
        x2=backimg.shape[1]
        usx=backimg.shape[1]-x1
        
    if y2>backimg.shape[0]:
        y2=backimg.shape[0]
        usy=backimg.shape[0]-y1
    
    
    if backimg.shape[2] == 3:       
        b, g, r = cv2.split(backimg) 
        alpha = np.ones(b.shape, dtype=b.dtype) * 255 # 创建Alpha通道 
        backimg = cv2.merge((b, g, r, alpha))
 
    alpha_huzip = mask[:usy,:usx,3] / 255.0
    
    #print("shapmask,mask3",alpha_huzip.shape,mask3[y1:y2, x1:x2].shape)
    #alpha_huzip=alpha_huzip
    mask[:usy,:usx,:3] = cv2.bitwise_and(mask[:usy,:usx,:3], mask[:usy,:usx,:3], mask = mask3[y1:y2, x1:x2])
    
    #(1-mask3[y1:y2, x1:x2])*backimg[y1:y2,x1:x2,c]    
    
    for cc in range(3):
        mask[:usy,:usx,cc]=mask[:usy,:usx,cc]*(1-dilated[y1:y2, x1:x2])+backimg[y1:y2,x1:x2,cc]*dilated[y1:y2, x1:x2]
   
    cv2.imshow("step3:",mask[:usy,:usx,:3])
    
    alpha_huzij = 1 - alpha_huzip
    
    
    some=0
    for i in range(y2-y1):
        for j in range(x2-x1):
            if mask3[i,j]!=0 or mask3[i,j]!=1:
                if mask3[i,j]>0.5:
                    mask3[i,j]=1
                else:
                    mask3[i,j]=0
    print("0,1:",some)
    
    maskal=1-mask3[y1:y2,x1:x2]
    ddratio=0.5
    for c in range(0,3):
        backimg[y1:y2, x1:x2, c] = (alpha_huzij*backimg[y1:y2,x1:x2,c]) + (alpha_huzip*((maskal*backimg[y1:y2,x1:x2,c]) + mask3[y1:y2,x1:x2]*((1-ddratio)*backimg[y1:y2,x1:x2,c]+ddratio*(mask[:usy,:usx,c]))))
        #backimg[y1:y2, x1:x2, c] = ((maskal*backimg[y1:y2,x1:x2,c]) + mask3[y1:y2,x1:x2]*((1-ddratio)*backimg[y1:y2,x1:x2,c]+ddratio*(mask[:usy,:usx,c])))
    
    #cv2.imshow('moni33tor', backimg[:,:,:3])
    return backimg[:,:,:3]

iscap=False
orang=cv2.imread('avatar1.png')
while True:
    # 读取摄像头画面
    if iscap:
        ret, frame = video_capture.read()
    else:
        frame=orang.copy()    

    # 改变摄像头图像的大小，图像小，所做的计算就少
    small_frame = cv2.resize(frame, (0, 0), fx=0.25, fy=0.25)

    # opencv的图像是BGR格式的，而我们需要是的RGB格式的，因此需要进行一个转换。
    rgb_small_frame = small_frame[:, :, ::-1]

    # Only process every other frame of video to save time
    if process_this_frame:
        # 根据encoding来判断是不是同一个人，是就输出true，不是为flase
     
        face_landmarks_list = face_recognition.face_landmarks(rgb_small_frame)
        for face_landmarks in face_landmarks_list:
        #打印此图像中每个面部特征的位置
            facial_features = [
                    
                    'chin',
                    'left_eyebrow',
                    'right_eyebrow',
                    'nose_bridge',
                    'nose_tip',
                    'left_eye',
                    'right_eye',
                    'top_lip',
                    'bottom_lip'
            ]

    

    #创建脸部遮罩
    mask1=np.zeros(frame.shape,np.uint8)
    #下脸部
    epoints=[]
    highestchin_y=10000
    leftchin_x=0
    rightchin_x=10000
    for point in face_landmarks["chin"]:
        epoints.append([point[0]*4,point[1]*4])  
        if highestchin_y>point[1]*4:
            highestchin_y=point[1]*4    
        if leftchin_x<point[0]*4:
            leftchin_x=point[0]*4
        if rightchin_x>point[0]*4:
            rightchin_x=point[0]*4   

    cv2.polylines(mask1, [np.array(epoints)], isClosed=True, color=[255, 255, 255], thickness=1)
    cv2.fillPoly(mask1, [np.array(epoints)], color=[255, 255, 255]) 
    
    uppoints=[[leftchin_x,highestchin_y],[rightchin_x,highestchin_y-10],[rightchin_x,0],[leftchin_x,0]]
    cv2.rectangle(mask1, (leftchin_x,0),(rightchin_x,highestchin_y+10), (255, 255, 255), -1)
   
    cv2.imshow("Step1:mask1",mask1)
    
    maskup=threshold(frame,midthreshold=88,maxthreshold=255,binarymod=cv2.THRESH_BINARY_INV)
    kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (7,7))
    dilated = cv2.dilate(maskup.copy(), kernel, 10)   
    
    cv2.imshow("Step2:",dilated)
    
    
   

    lef_center_x=0
    lef_center_y=0
    epoints=[]
    for point in face_landmarks["left_eye"]:
        epoints.append([point[0]*4,point[1]*4])    

            
    cv2.polylines(mask1, [np.array(epoints)], isClosed=True, color=[0, 0, 255], thickness=5)
    cv2.fillPoly(mask1, [np.array(epoints)], color=[0, 0, 0])

    rig_center_x=0
    rig_center_y=0
    epoints=[]
    for point in face_landmarks["right_eye"]:
            epoints.append([point[0]*4,point[1]*4])    
    cv2.polylines(mask1, [np.array(epoints)], isClosed=True, color=[0, 0, 255], thickness=5)
    cv2.fillPoly(mask1, [np.array(epoints)], color=[0, 0, 0])


    top_lip_x=0
    top_lip_y=0
    epoints=[]
    for point in face_landmarks["top_lip"]:
        top_lip_x+=point[0]*4
        top_lip_y+=point[1]*4
        epoints.append([point[0]*4,point[1]*4]) 
    
    top_lip_x= int(top_lip_x/len(face_landmarks["top_lip"]))   
    top_lip_y= int(top_lip_y/len(face_landmarks["top_lip"]))      
   
    cv2.polylines(mask1, [np.array(epoints)], isClosed=True, color=[0, 0, 0], thickness=1)
    cv2.fillPoly(mask1, [np.array(epoints)], color=[0, 0, 0])  

   
    epoints=[]
    for point in face_landmarks["bottom_lip"]:
        epoints.append([point[0]*4,point[1]*4])           
   
    cv2.polylines(mask1, [np.array(epoints)], isClosed=True, color=[0, 0, 0], thickness=1)
    cv2.fillPoly(mask1, [np.array(epoints)], color=[0, 0, 0])  
   
    
    maskgray = cv2.cvtColor(mask1,cv2.COLOR_BGR2GRAY)
    ret,mask = cv2.threshold(maskgray,175,255,cv2.THRESH_BINARY)

    
    lef_center_x=0
    lef_center_y=0
    
    for point in face_landmarks["left_eye"]:
            lef_center_x+=point[0]*4
            lef_center_y+=point[1]*4
    lef_center_x=int(lef_center_x/len(face_landmarks["left_eye"]))
    lef_center_y=int(lef_center_y/len(face_landmarks["left_eye"]))
    
    rig_center_x=0
    rig_center_y=0
    
    for point in face_landmarks["right_eye"]:
            rig_center_x+=point[0]*4
            rig_center_y+=point[1]*4
    rig_center_x=int(rig_center_x/len(face_landmarks["left_eye"]))
    rig_center_y=int(rig_center_y/len(face_landmarks["left_eye"]))
    
    
    
    dilated[highestchin_y:,:]=0
    
    cv2.circle(dilated, (lef_center_x,lef_center_y), 105, 0, -1)  
    cv2.circle(dilated, (rig_center_x,rig_center_y), 105, 0, -1)     
    
    for i in range(dilated.shape[0]):
        for j in range(dilated.shape[1]):
            if dilated[i,j]>0.5:
                dilated[i,j]=1
            else:
                dilated[i,j]=0
                
        
    
    cv2.imshow("d",dilated)
    #去掉眼睛和嘴唇
    
    
    nosey=0
    for point in face_landmarks["nose_tip"]:
        if nosey<point[1]*4:
            nosey=point[1]*4
        
    print("nosey",nosey,nosey-lef_center_y)
    output1=putonmask(maskk,frame,mask,dilated,(rig_center_x-lef_center_x),(nosey-lef_center_y),leftcenter=(lef_center_x,lef_center_y))
    cv2.imshow('monitor', output1)
    time.sleep(0.1)
    # 按Q退出
    if cv2.waitKey(1) & 0xFF == ord('q'):
        break

video_capture.release()
cv2.destroyAllWindows()

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
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271

代码用到的原图如下：