笑脸识别-dlib
import cv2 # 图像处理的库 OpenCvimport dlib# 人脸识别的库 dlibimport numpy as np # 数据处理的库 numpyclass face_emotion():def __init__(self):self.detector = dlib.get_frontal_face_detector()self.predictor = dlib.shape_predictor("shape_predictor_68_face_landmarks.dat")self.cap = cv2.VideoCapture(0)self.cap.set(3, 480)t = 0 def learning_face(self):line_brow_x = []line_brow_y = []while(self.cap.isOpened()):flag, im_rd = self.cap.read()k = cv2.waitKey(1)# 取灰度img_gray = cv2.cvtColor(im_rd, cv2.COLOR_RGB2GRAY) faces = self.detector(img_gray, 0)font = cv2.FONT_HERSHEY_SIMPLEX# 如果检测到人脸if(len(faces) != 0):# 对每个人脸都标出68个特征点for i in range(len(faces)):for k, d in enumerate(faces):cv2.rectangle(im_rd, (d.left(), d.top()), (d.right(), d.bottom()), (0,0,255))self.face_width = d.right() - d.left()shape = self.predictor(im_rd, d)mouth_width = (shape.part(54).x - shape.part(48).x) / self.face_width mouth_height = (shape.part(66).y - shape.part(62).y) / self.face_widthbrow_sum = 0 frown_sum = 0 for j in range(17, 21):brow_sum += (shape.part(j).y - d.top()) + (shape.part(j + 5).y - d.top())frown_sum += shape.part(j + 5).x - shape.part(j).xline_brow_x.append(shape.part(j).x)line_brow_y.append(shape.part(j).y)tempx = np.array(line_brow_x)tempy = np.array(line_brow_y)z1 = np.polyfit(tempx, tempy, 1) self.brow_k = -round(z1[0], 3) brow_height = (brow_sum / 10) / self.face_width # 眉毛高度占比brow_width = (frown_sum / 5) / self.face_width # 眉毛距离占比eye_sum = (shape.part(41).y - shape.part(37).y + shape.part(40).y - shape.part(38).y + shape.part(47).y - shape.part(43).y + shape.part(46).y - shape.part(44).y)eye_hight = (eye_sum / 4) / self.face_widthif round(mouth_height >= 0.03) and eye_hight<0.56:cv2.putText(im_rd, "smile", (d.left(), d.bottom() + 20), cv2.FONT_HERSHEY_SIMPLEX, 2,(0,255,0), 2, 4)if round(mouth_height<0.03) and self.brow_k>-0.3:cv2.putText(im_rd, "unsmile", (d.left(), d.bottom() + 20), cv2.FONT_HERSHEY_SIMPLEX, 2,(0,255,0), 2, 4)cv2.putText(im_rd, "Face-" + str(len(faces)), (20,50), font, 0.6, (0,0,255), 1, cv2.LINE_AA)else:cv2.putText(im_rd, "No Face", (20,50), font, 0.6, (0,0,255), 1, cv2.LINE_AA)im_rd = cv2.putText(im_rd, "S: screenshot", (20,450), font, 0.6, (255,0,255), 1, cv2.LINE_AA)im_rd = cv2.putText(im_rd, "Q: quit", (20,470), font, 0.6, (255,0,255), 1, cv2.LINE_AA)if (cv2.waitKey(1) & 0xFF) == ord('s'):t += 1cv2.imwrite("screenshoot" + str(t) + ".jpg", im_rd)# 按下 q 键退出if (cv2.waitKey(1)) == ord('q'):break# 窗口显示cv2.imshow("Face Recognition", im_rd)self.cap.release()cv2.destroyAllWindows()if __name__ == "__main__":my_face = face_emotion()my_face.learning_face()
结果如下：
笑脸识别-CNN
图片预处理
import dlib # 人脸识别的库dlibimport numpy as np # 数据处理的库numpyimport cv2# 图像处理的库OpenCvimport os# dlib预测器detector = dlib.get_frontal_face_detector()predictor = dlib.shape_predictor('shape_predictor_68_face_landmarks.dat')# 读取图像的路径path_read = "files"for file_name in os.listdir(path_read):#aa是图片的全路径aa=(path_read +"/"+file_name)#读入的图片的路径中含非英文img=cv2.imdecode(np.fromfile(aa, dtype=np.uint8), cv2.IMREAD_UNCHANGED)#获取图片的宽高img_shape=img.shapeimg_height=img_shape[0]img_width=img_shape[1]# 用来存储生成的单张人脸的路径path_save="files1" # dlib检测dets = detector(img,1)print("人脸数：", len(dets))for k, d in enumerate(dets):if len(dets)>1:continue# 计算矩形大小# (x,y), (宽度width, 高度height)pos_start = tuple([d.left(), d.top()])pos_end = tuple([d.right(), d.bottom()])# 计算矩形框大小height = d.bottom()-d.top()width = d.right()-d.left()# 根据人脸大小生成空的图像img_blank = np.zeros((height, width, 3), np.uint8)for i in range(height):if d.top()+i>=img_height:# 防止越界continuefor j in range(width):if d.left()+j>=img_width:# 防止越界continueimg_blank[i][j] = img[d.top()+i][d.left()+j]img_blank = cv2.resize(img_blank, (200, 200), interpolation=cv2.INTER_CUBIC)cv2.imencode('.jpg', img_blank)[1].tofile(path_save+"/"+file_name) # 正确方法
划分数据集
方法一：
import os, shutil# 原始数据集路径original_dataset_dir = 'files1'# 新的数据集base_dir = 'files2'os.mkdir(base_dir)# 训练图像、验证图像、测试图像的目录train_dir = os.path.join(base_dir, 'train')os.mkdir(train_dir)validation_dir = os.path.join(base_dir, 'validation')os.mkdir(validation_dir)test_dir = os.path.join(base_dir, 'test')os.mkdir(test_dir)train_cats_dir = os.path.join(train_dir, 'smile')os.mkdir(train_c_dir)train_dogs_dir = os.path.join(train_dir, 'unsmile')os.mkdir(train_d_dir)validation_cats_dir = os.path.join(validation_dir, 'smile')os.mkdir(validation_c_dir)validation_dogs_dir = os.path.join(validation_dir, 'unsmile')os.mkdir(validation_d_dir)test_cats_dir = os.path.join(test_dir, 'smile')os.mkdir(test_c_dir)test_dogs_dir = os.path.join(test_dir, 'unsmile')os.mkdir(test_d_dir)# 复制1000张笑脸图片到train_c_dirfnames = ['file{}.jpg'.format(i) for i in range(900)]for fname in fnames:src = os.path.join(original_dataset_dir, fname)dst = os.path.join(train_c_dir, fname)shutil.copyfile(src, dst)fnames = ['file{}.jpg'.format(i) for i in range(900, 1350)]for fname in fnames:src = os.path.join(original_dataset_dir, fname)dst = os.path.join(validation_c_dir, fname)shutil.copyfile(src, dst)# Copy next 500 cat images to test_cats_dirfnames = ['file{}.jpg'.format(i) for i in range(1350, 1800)]for fname in fnames:src = os.path.join(original_dataset_dir, fname)dst = os.path.join(test_c_dir, fname)shutil.copyfile(src, dst)fnames = ['file{}.jpg'.format(i) for i in range(900)]for fname in fnames:src = os.path.join(original_dataset_dir, fname)dst = os.path.join(train_d_dir, fname)shutil.copyfile(src, dst)# Copy next 500 dog images to validation_dogs_dirfnames = ['file{}.jpg'.format(i) for i in range(900, 1350)]for fname in fnames:src = os.path.join(original_dataset_dir, fname)dst = os.path.join(validation_d_dir, fname)shutil.copyfile(src, dst)# Copy next 500 dog images to test_dogs_dirfnames = ['file{}.jpg'.format(i) for i in range(1350, 1800)]for fname in fnames:src = os.path.join(original_dataset_dir, fname)dst = os.path.join(test_d_dir, fname)shutil.copyfile(src, dst)
方法二
import kerasimport os, shutiltrain_smile_dir="files2/train/smile/"train_umsmile_dir="files2/train/unsmile/"test_smile_dir="files2/test/smile/"test_umsmile_dir="files2/test/unsmile/"validation_smile_dir="files2/validation/smile/"validation_unsmile_dir="files2/validation/unsmile/"train_dir="files2/train/"test_dir="files2/test/"validation_dir="files2/validation/"
查看文件夹下图片的数量：
print('total training smile images:', len(os.listdir(train_smile_dir)))print('total training unsmile images:', len(os.listdir(train_umsmile_dir)))print('total testing smile images:', len(os.listdir(test_smile_dir)))print('total testing unsmile images:', len(os.listdir(test_umsmile_dir)))print('total validation smile images:', len(os.listdir(validation_smile_dir)))print('total validation unsmile images:', len(os.listdir(validation_unsmile_dir)))
创建模型
#创建模型from keras import layersfrom keras import modelsmodel = models.Sequential()model.add(layers.Conv2D(32, (3, 3), activation='relu',input_shape=(150, 150, 3)))model.add(layers.MaxPooling2D((2, 2)))model.add(layers.Conv2D(64, (3, 3), activation='relu'))model.add(layers.MaxPooling2D((2, 2)))model.add(layers.Conv2D(128, (3, 3), activation='relu'))model.add(layers.MaxPooling2D((2, 2)))model.add(layers.Conv2D(128, (3, 3), activation='relu'))model.add(layers.MaxPooling2D((2, 2)))model.add(layers.Flatten())model.add(layers.Dense(512, activation='relu'))model.add(layers.Dense(1, activation='sigmoid'))
查看模型：
model.summary()
归一化处理
#归一化from keras import pile(loss='binary_crossentropy',optimizer=optimizers.RMSprop(lr=1e-4),metrics=['acc'])
from keras.preprocessing.image import ImageDataGeneratortrain_datagen = ImageDataGenerator(rescale=1./255)validation_datagen=ImageDataGenerator(rescale=1./255)test_datagen = ImageDataGenerator(rescale=1./255)train_generator = train_datagen.flow_from_directory(# 目标文件目录train_dir,#所有图片的size必须是150x150target_size=(150, 150),batch_size=20,# Since we use binary_crossentropy loss, we need binary labelsclass_mode='binary')validation_generator = test_datagen.flow_from_directory(validation_dir,target_size=(150, 150),batch_size=20,class_mode='binary')test_generator = test_datagen.flow_from_directory(test_dir,target_size=(150, 150),batch_size=20,class_mode='binary')
数据增强
#数据增强datagen = ImageDataGenerator(rotation_range=40,width_shift_range=0.2,height_shift_range=0.2,shear_range=0.2,zoom_range=0.2,horizontal_flip=True,fill_mode='nearest')
#数据增强后图片变化import matplotlib.pyplot as plt# This is module with image preprocessing utilitiesfrom keras.preprocessing import imagefnames = [os.path.join(train_smile_dir, fname) for fname in os.listdir(train_smile_dir)]img_path = fnames[3]img = image.load_img(img_path, target_size=(150, 150))x = image.img_to_array(img)x = x.reshape((1,) + x.shape)i = 0for batch in datagen.flow(x, batch_size=1):plt.figure(i)imgplot = plt.imshow(image.array_to_img(batch[0]))i += 1if i % 4 == 0:breakplt.show()
结果：
创建网络
#创建网络model = models.Sequential()model.add(layers.Conv2D(32, (3, 3), activation='relu',input_shape=(150, 150, 3)))model.add(layers.MaxPooling2D((2, 2)))model.add(layers.Conv2D(64, (3, 3), activation='relu'))model.add(layers.MaxPooling2D((2, 2)))model.add(layers.Conv2D(128, (3, 3), activation='relu'))model.add(layers.MaxPooling2D((2, 2)))model.add(layers.Conv2D(128, (3, 3), activation='relu'))model.add(layers.MaxPooling2D((2, 2)))model.add(layers.Flatten())model.add(layers.Dropout(0.5))model.add(layers.Dense(512, activation='relu'))model.add(layers.Dense(1, activation='sigmoid'))pile(loss='binary_crossentropy',optimizer=optimizers.RMSprop(lr=1e-4),metrics=['acc'])
归一化处理：
#归一化处理train_datagen = ImageDataGenerator(rescale=1./255,rotation_range=40,width_shift_range=0.2,height_shift_range=0.2,shear_range=0.2,zoom_range=0.2,horizontal_flip=True,)test_datagen = ImageDataGenerator(rescale=1./255)train_generator = train_datagen.flow_from_directory(# This is the target directorytrain_dir,# All images will be resized to 150x150target_size=(150, 150),batch_size=32,# Since we use binary_crossentropy loss, we need binary labelsclass_mode='binary')validation_generator = test_datagen.flow_from_directory(validation_dir,target_size=(150, 150),batch_size=32,class_mode='binary')history = model.fit_generator(train_generator,steps_per_epoch=100,epochs=60, validation_data=validation_generator,validation_steps=50)
保存模型：
model.save('smileAndUnsmile1.h5')
结果：
单张图片测试
# 单张图片进行判断 是笑脸还是非笑脸import cv2from keras.preprocessing import imagefrom keras.models import load_modelimport numpy as np#加载模型model = load_model('smileAndUnsmile1.h5')#本地图片路径img_path='test.jpg'img = image.load_img(img_path, target_size=(150, 150))img_tensor = image.img_to_array(img)/255.0img_tensor = np.expand_dims(img_tensor, axis=0)prediction =model.predict(img_tensor) print(prediction)if prediction[0][0]>0.5:result='非笑脸'else:result='笑脸'print(result)
结果如下：
摄像头测试
#检测视频或者摄像头中的人脸import cv2from keras.preprocessing import imagefrom keras.models import load_modelimport numpy as npimport dlibfrom PIL import Imagemodel = load_model('smileAndUnsmile1.h5')detector = dlib.get_frontal_face_detector()video=cv2.VideoCapture(0)font = cv2.FONT_HERSHEY_SIMPLEXdef rec(img):gray=cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)dets=detector(gray,1)if dets is not None:for face in dets:left=face.left()top=face.top()right=face.right()bottom=face.bottom()cv2.rectangle(img,(left,top),(right,bottom),(0,255,0),2)img1=cv2.resize(img[top:bottom,left:right],dsize=(150,150))img1=cv2.cvtColor(img1,cv2.COLOR_BGR2RGB)img1 = np.array(img1)/255.img_tensor = img1.reshape(-1,150,150,3)prediction =model.predict(img_tensor) if prediction[0][0]>0.5:result='unsmile'else:result='smile'cv2.putText(img, result, (left,top), font, 2, (0, 255, 0), 2, cv2.LINE_AA)cv2.imshow('Video', img)while video.isOpened():res, img_rd = video.read()if not res:breakrec(img_rd)if cv2.waitKey(1) & 0xFF == ord('q'):breakvideo.release()cv2.destroyAllWindows()
结果如下：
口罩识别
图片预处理
import dlib # 人脸识别的库dlibimport numpy as np # 数据处理的库numpyimport cv2# 图像处理的库OpenCvimport os# dlib预测器detector = dlib.get_frontal_face_detector()predictor = dlib.shape_predictor('shape_predictor_68_face_landmarks.dat')# 读取图像的路径path_read = "data"for file_name in os.listdir(path_read):#aa是图片的全路径aa=(path_read +"/"+file_name)#读入的图片的路径中含非英文img=cv2.imdecode(np.fromfile(aa, dtype=np.uint8), cv2.IMREAD_UNCHANGED)#获取图片的宽高img_shape=img.shapeimg_height=img_shape[0]img_width=img_shape[1]# 用来存储生成的单张人脸的路径path_save="maskdata" # dlib检测dets = detector(img,1)print("人脸数：", len(dets))for k, d in enumerate(dets):if len(dets)>1:continue# 计算矩形大小# (x,y), (宽度width, 高度height)pos_start = tuple([d.left(), d.top()])pos_end = tuple([d.right(), d.bottom()])# 计算矩形框大小height = d.bottom()-d.top()width = d.right()-d.left()# 根据人脸大小生成空的图像img_blank = np.zeros((height, width, 3), np.uint8)for i in range(height):if d.top()+i>=img_height:# 防止越界continuefor j in range(width):if d.left()+j>=img_width:# 防止越界continueimg_blank[i][j] = img[d.top()+i][d.left()+j]img_blank = cv2.resize(img_blank, (200, 200), interpolation=cv2.INTER_CUBIC)cv2.imencode('.jpg', img_blank)[1].tofile(path_save+"/"+file_name) # 正确方法
导入数据集
import kerasimport os, shutiltrain_havemask_dir="maskdata/train/mask/"train_nomask_dir="maskdata/train/nomask/"test_havemask_dir="maskdata/test/mask/"test_nomask_dir="maskdata/test/nomask/"validation_havemask_dir="maskdata/validation/mask/"validation_nomask_dir="maskdata/validation/nomask/"train_dir="maskdata/train/"test_dir="maskdata/test/"validation_dir="maskdata/validation/"
创建模型
#创建模型from keras import layersfrom keras import modelsmodel = models.Sequential()model.add(layers.Conv2D(32, (3, 3), activation='relu',input_shape=(150, 150, 3)))model.add(layers.MaxPooling2D((2, 2)))model.add(layers.Conv2D(64, (3, 3), activation='relu'))model.add(layers.MaxPooling2D((2, 2)))model.add(layers.Conv2D(128, (3, 3), activation='relu'))model.add(layers.MaxPooling2D((2, 2)))model.add(layers.Conv2D(128, (3, 3), activation='relu'))model.add(layers.MaxPooling2D((2, 2)))model.add(layers.Flatten())model.add(layers.Dense(512, activation='relu'))model.add(layers.Dense(1, activation='sigmoid'))
查看模型：
model.summary()
归一化处理
from keras import pile(loss='binary_crossentropy',optimizer=optimizers.RMSprop(lr=1e-4),metrics=['acc'])
from keras.preprocessing.image import ImageDataGenerator# All images will be rescaled by 1./255train_datagen = ImageDataGenerator(rescale=1./255)validation_datagen=ImageDataGenerator(rescale=1./255)test_datagen = ImageDataGenerator(rescale=1./255)train_generator = train_datagen.flow_from_directory(# 目标文件目录train_dir,#所有图片的size必须是150x150target_size=(150, 150),batch_size=20,# Since we use binary_crossentropy loss, we need binary labelsclass_mode='binary')validation_generator = test_datagen.flow_from_directory(validation_dir,target_size=(150, 150),batch_size=20,class_mode='binary')test_generator = test_datagen.flow_from_directory(test_dir,target_size=(150, 150),batch_size=20,class_mode='binary')
for data_batch, labels_batch in train_generator:print('data batch shape:', data_batch.shape)print('labels batch shape:', labels_batch)break
训练模型
#耗时长history = model.fit_generator(train_generator,steps_per_epoch=100,epochs=10,validation_data=validation_generator,validation_steps=50)
保存模型：
model.save('maskAndNomask1.h5')
数据增强
#数据增强datagen = ImageDataGenerator(rotation_range=40,width_shift_range=0.2,height_shift_range=0.2,shear_range=0.2,zoom_range=0.2,horizontal_flip=True,fill_mode='nearest')
数据增强前后对比：
import matplotlib.pyplot as pltfrom keras.preprocessing import imagefnames = [os.path.join(train_havemask_dir, fname) for fname in os.listdir(train_havemask_dir)]img_path = fnames[3]img = image.load_img(img_path, target_size=(150, 150))x = image.img_to_array(img)x = x.reshape((1,) + x.shape)i = 0for batch in datagen.flow(x, batch_size=1):plt.figure(i)imgplot = plt.imshow(image.array_to_img(batch[0]))i += 1if i % 4 == 0:breakplt.show()
结果如下：
创建网络
model = models.Sequential()model.add(layers.Conv2D(32, (3, 3), activation='relu',input_shape=(150, 150, 3)))model.add(layers.MaxPooling2D((2, 2)))model.add(layers.Conv2D(64, (3, 3), activation='relu'))model.add(layers.MaxPooling2D((2, 2)))model.add(layers.Conv2D(128, (3, 3), activation='relu'))model.add(layers.MaxPooling2D((2, 2)))model.add(layers.Conv2D(128, (3, 3), activation='relu'))model.add(layers.MaxPooling2D((2, 2)))model.add(layers.Flatten())model.add(layers.Dropout(0.5))model.add(layers.Dense(512, activation='relu'))model.add(layers.Dense(1, activation='sigmoid'))pile(loss='binary_crossentropy',optimizer=optimizers.RMSprop(lr=1e-4),metrics=['acc'])
归一化处理，代码如下：
#归一化处理train_datagen = ImageDataGenerator(rescale=1./255,rotation_range=40,width_shift_range=0.2,height_shift_range=0.2,shear_range=0.2,zoom_range=0.2,horizontal_flip=True,)test_datagen = ImageDataGenerator(rescale=1./255)train_generator = train_datagen.flow_from_directory(# This is the target directorytrain_dir,# All images will be resized to 150x150target_size=(150, 150),batch_size=32,# Since we use binary_crossentropy loss, we need binary labelsclass_mode='binary')validation_generator = test_datagen.flow_from_directory(validation_dir,target_size=(150, 150),batch_size=32,class_mode='binary')history = model.fit_generator(train_generator,steps_per_epoch=100,epochs=60, validation_data=validation_generator,validation_steps=50)
保存模型：
model.save('maskAndNomask2.h5')
单张图片测试
# 单张图片进行判断 是否戴口罩import cv2from keras.preprocessing import imagefrom keras.models import load_modelimport numpy as npmodel = load_model('maskAndNomask2.h5')img_path='masktest.jpg'img = image.load_img(img_path, target_size=(150, 150))#print(img.size)img_tensor = image.img_to_array(img)/255.0img_tensor = np.expand_dims(img_tensor, axis=0)prediction =model.predict(img_tensor) print(prediction)if prediction[0][0]>0.5:result='未戴口罩'else:result='戴口罩'print(result)
结果如下：
摄像头测试
import cv2from keras.preprocessing import imagefrom keras.models import load_modelimport numpy as npimport dlibfrom PIL import Imagemodel = load_model('maskAndNomask2.h5')detector = dlib.get_frontal_face_detector()# video=cv2.VideoCapture('media/video.mp4')# video=cv2.VideoCapture('data/face_recognition.mp4')video=cv2.VideoCapture(0)font = cv2.FONT_HERSHEY_SIMPLEXdef rec(img):gray=cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)dets=detector(gray,1)if dets is not None:for face in dets:left=face.left()top=face.top()right=face.right()bottom=face.bottom()cv2.rectangle(img,(left,top),(right,bottom),(0,255,0),2)def mask(img):img1=cv2.resize(img,dsize=(150,150))img1=cv2.cvtColor(img1,cv2.COLOR_BGR2RGB)img1 = np.array(img1)/255.img_tensor = img1.reshape(-1,150,150,3)prediction =model.predict(img_tensor) if prediction[0][0]>0.5:result='no-mask'else:result='have-mask'cv2.putText(img, result, (100,200), font, 2, (0, 255, 0), 2, cv2.LINE_AA)cv2.imshow('Video', img)while video.isOpened():res, img_rd = video.read()if not res:break#将视频每一帧传入两个函数，分别用于圈出人脸与判断是否带口罩rec(img_rd)mask(img_rd)#q关闭窗口if cv2.waitKey(1) & 0xFF == ord('q'):breakvideo.release()cv2.destroyAllWindows()
结果如下：
