Python实现人脸识别算法并封装为类库

引言

人脸识别技术在现代社会中应用广泛，从安全监控到智能门锁，再到社交媒体中的照片标记功能，都离不开这项技术。本文将详细介绍如何使用Python实现基本的人脸识别算法，并将其封装为一个类库，以便在多个项目中复用。

环境准备

安装依赖

首先，确保你的环境中安装了以下库：

• OpenCV
• dlib
• face_recognition

你可以使用以下命令安装这些库：

pip install opencv-python
pip install dlib
pip install face_recognition

代码实现

1. 创建类库文件

我们将创建一个名为 face_recognition_lib.py 的文件，内容如下：

import cv2
import face_recognition
import numpy as np
from PIL import Image, ImageDrawclass FaceRecognition:def __init__(self, image_paths):"""初始化FaceRecognition类。:param image_paths: 已知人脸图像的路径列表"""self.known_face_encodings, self.known_face_names = self.load_and_encode_faces(image_paths)def load_and_encode_faces(self, image_paths):"""加载并编码已知人脸图像。:param image_paths: 已知人脸图像的路径列表:return: 已知人脸的编码和名称"""known_face_encodings = []known_face_names = []for image_path in image_paths:# 加载图片image = face_recognition.load_image_file(image_path)# 编码人脸face_encoding = face_recognition.face_encodings(image)[0]# 获取文件名作为名字name = image_path.split('/')[-1].split('.')[0]# 添加到已知人脸列表known_face_encodings.append(face_encoding)known_face_names.append(name)return known_face_encodings, known_face_namesdef real_time_face_recognition(self):"""实现实时人脸识别。"""# 打开摄像头video_capture = cv2.VideoCapture(0)while True:# 读取一帧ret, frame = video_capture.read()# 将帧转换为RGBrgb_frame = frame[:, :, ::-1]# 检测人脸位置face_locations = face_recognition.face_locations(rgb_frame)# 编码人脸face_encodings = face_recognition.face_encodings(rgb_frame, face_locations)# 遍历检测到的人脸for (top, right, bottom, left), face_encoding in zip(face_locations, face_encodings):# 匹配已知人脸matches = face_recognition.compare_faces(self.known_face_encodings, face_encoding)name = "Unknown"# 计算欧氏距离face_distances = face_recognition.face_distance(self.known_face_encodings, face_encoding)best_match_index = np.argmin(face_distances)if matches[best_match_index]:name = self.known_face_names[best_match_index]# 在帧上绘制矩形和标签cv2.rectangle(frame, (left, top), (right, bottom), (0, 0, 255), 2)cv2.rectangle(frame, (left, bottom - 35), (right, bottom), (0, 0, 255), cv2.FILLED)font = cv2.FONT_HERSHEY_DUPLEXcv2.putText(frame, name, (left + 6, bottom - 6), font, 1.0, (255, 255, 255), 1)# 显示结果cv2.imshow('Video', frame)# 按q键退出if cv2.waitKey(1) & 0xFF == ord('q'):break# 释放摄像头video_capture.release()cv2.destroyAllWindows()# 示例用法
if __name__ == "__main__":# 已知人脸图像路径image_paths = ['path/to/known_face_1.jpg','path/to/known_face_2.jpg',# 添加更多已知人脸图像路径]# 初始化FaceRecognition类face_recognition = FaceRecognition(image_paths)# 启动实时人脸识别face_recognition.real_time_face_recognition()

2. 详细解释

初始化

class FaceRecognition:def __init__(self, image_paths):"""初始化FaceRecognition类。:param image_paths: 已知人脸图像的路径列表"""self.known_face_encodings, self.known_face_names = self.load_and_encode_faces(image_paths)

在类的初始化方法中，我们传入已知人脸图像的路径列表，并调用 load_and_encode_faces 方法来加载和编码这些图像。

加载和编码已知人脸

def load_and_encode_faces(self, image_paths):"""加载并编码已知人脸图像。:param image_paths: 已知人脸图像的路径列表:return: 已知人脸的编码和名称"""known_face_encodings = []known_face_names = []for image_path in image_paths:# 加载图片image = face_recognition.load_image_file(image_path)# 编码人脸face_encoding = face_recognition.face_encodings(image)[0]# 获取文件名作为名字name = image_path.split('/')[-1].split('.')[0]# 添加到已知人脸列表known_face_encodings.append(face_encoding)known_face_names.append(name)return known_face_encodings, known_face_names

load_and_encode_faces 方法遍历传入的图像路径列表，加载每张图像并对其进行编码。编码后的特征向量和对应的名字被存储在两个列表中，分别返回。

实时人脸识别

def real_time_face_recognition(self):"""实现实时人脸识别。"""# 打开摄像头video_capture = cv2.VideoCapture(0)while True:# 读取一帧ret, frame = video_capture.read()# 将帧转换为RGBrgb_frame = frame[:, :, ::-1]# 检测人脸位置face_locations = face_recognition.face_locations(rgb_frame)# 编码人脸face_encodings = face_recognition.face_encodings(rgb_frame, face_locations)# 遍历检测到的人脸for (top, right, bottom, left), face_encoding in zip(face_locations, face_encodings):# 匹配已知人脸matches = face_recognition.compare_faces(self.known_face_encodings, face_encoding)name = "Unknown"# 计算欧氏距离face_distances = face_recognition.face_distance(self.known_face_encodings, face_encoding)best_match_index = np.argmin(face_distances)if matches[best_match_index]:name = self.known_face_names[best_match_index]# 在帧上绘制矩形和标签cv2.rectangle(frame, (left, top), (right, bottom), (0, 0, 255), 2)cv2.rectangle(frame, (left, bottom - 35), (right, bottom), (0, 0, 255), cv2.FILLED)font = cv2.FONT_HERSHEY_DUPLEXcv2.putText(frame, name, (left + 6, bottom - 6), font, 1.0, (255, 255, 255), 1)# 显示结果cv2.imshow('Video', frame)# 按q键退出if cv2.waitKey(1) & 0xFF == ord('q'):break# 释放摄像头video_capture.release()cv2.destroyAllWindows()

real_time_face_recognition 方法打开摄像头并读取视频流。对于每一帧，我们将其转换为RGB格式，检测人脸位置并编码。然后，我们将编码后的人脸与已知人脸进行匹配，并在帧上绘制矩形和标签。按 q 键可以退出程序。

3. 使用类库

创建一个名为 main.py 的文件，内容如下：

from face_recognition_lib import FaceRecognition# 已知人脸图像路径
image_paths = ['path/to/known_face_1.jpg','path/to/known_face_2.jpg',# 添加更多已知人脸图像路径
]# 初始化FaceRecognition类
face_recognition = FaceRecognition(image_paths)# 启动实时人脸识别
face_recognition.real_time_face_recognition()

4. 运行示例

确保你的环境中安装了所需的库：

pip install opencv-python
pip install dlib
pip install face_recognition

然后运行 main.py：

python main.py

5. 代码运行接口

为了使类库更加通用和易于集成，我们可以添加一些额外的方法和配置选项。以下是改进后的类库文件 face_recognition_lib.py：

import cv2
import face_recognition
import numpy as np
from PIL import Image, ImageDrawclass FaceRecognition:def __init__(self, image_paths, tolerance=0.6, model='hog'):"""初始化FaceRecognition类。:param image_paths: 已知人脸图像的路径列表:param tolerance: 人脸匹配的阈值:param model: 人脸检测模型（'hog' 或 'cnn'）"""self.tolerance = toleranceself.model = modelself.known_face_encodings, self.known_face_names = self.load_and_encode_faces(image_paths)def load_and_encode_faces(self, image_paths):"""加载并编码已知人脸图像。:param image_paths: 已知人脸图像的路径列表:return: 已知人脸的编码和名称"""known_face_encodings = []known_face_names = []for image_path in image_paths:# 加载图片image = face_recognition.load_image_file(image_path)# 编码人脸face_encoding = face_recognition.face_encodings(image)[0]# 获取文件名作为名字name = image_path.split('/')[-1].split('.')[0]# 添加到已知人脸列表known_face_encodings.append(face_encoding)known_face_names.append(name)return known_face_encodings, known_face_namesdef real_time_face_recognition(self, camera_index=0):"""实现实时人脸识别。:param camera_index: 摄像头索引"""# 打开摄像头video_capture = cv2.VideoCapture(camera_index)while True:# 读取一帧ret, frame = video_capture.read()# 将帧转换为RGBrgb_frame = frame[:, :, ::-1]# 检测人脸位置face_locations = face_recognition.face_locations(rgb_frame, model=self.model)# 编码人脸face_encodings = face_recognition.face_encodings(rgb_frame, face_locations)# 遍历检测到的人脸for (top, right, bottom, left), face_encoding in zip(face_locations, face_encodings):# 匹配已知人脸matches = face_recognition.compare_faces(self.known_face_encodings, face_encoding, tolerance=self.tolerance)name = "Unknown"# 计算欧氏距离face_distances = face_recognition.face_distance(self.known_face_encodings, face_encoding)best_match_index = np.argmin(face_distances)if matches[best_match_index]:name = self.known_face_names[best_match_index]# 在帧上绘制矩形和标签cv2.rectangle(frame, (left, top), (right, bottom), (0, 0, 255), 2)cv2.rectangle(frame, (left, bottom - 35), (right, bottom), (0, 0, 255), cv2.FILLED)font = cv2.FONT_HERSHEY_DUPLEXcv2.putText(frame, name, (left + 6, bottom - 6), font, 1.0, (255, 255, 255), 1)# 显示结果cv2.imshow('Video', frame)# 按q键退出if cv2.waitKey(1) & 0xFF == ord('q'):break# 释放摄像头video_capture.release()cv2.destroyAllWindows()# 示例用法
if __name__ == "__main__":# 已知人脸图像路径image_paths = ['path/to/known_face_1.jpg','path/to/known_face_2.jpg',# 添加更多已知人脸图像路径]# 初始化FaceRecognition类face_recognition = FaceRecognition(image_paths, tolerance=0.6, model='hog')# 启动实时人脸识别face_recognition.real_time_face_recognition(camera_index=0)

6. 详细说明

初始化

def __init__(self, image_paths, tolerance=0.6, model='hog'):"""初始化FaceRecognition类。:param image_paths: 已知人脸图像的路径列表:param tolerance: 人脸匹配的阈值:param model: 人脸检测模型（'hog' 或 'cnn'）"""self.tolerance = toleranceself.model = modelself.known_face_encodings, self.known_face_names = self.load_and_encode_faces(image_paths)

我们在初始化方法中增加了 tolerance 和 model 参数，以便在实例化类时可以调整人脸匹配的阈值和检测模型。

实时人脸识别

def real_time_face_recognition(self, camera_index=0):"""实现实时人脸识别。:param camera_index: 摄像头索引"""# 打开摄像头video_capture = cv2.VideoCapture(camera_index)while True:# 读取一帧ret, frame = video_capture.read()# 将帧转换为RGBrgb_frame = frame[:, :, ::-1]# 检测人脸位置face_locations = face_recognition.face_locations(rgb_frame, model=self.model)# 编码人脸face_encodings = face_recognition.face_encodings(rgb_frame, face_locations)# 遍历检测到的人脸for (top, right, bottom, left), face_encoding in zip(face_locations, face_encodings):# 匹配已知人脸matches = face_recognition.compare_faces(self.known_face_encodings, face_encoding, tolerance=self.tolerance)name = "Unknown"# 计算欧氏距离face_distances = face_recognition.face_distance(self.known_face_encodings, face_encoding)best_match_index = np.argmin(face_distances)if matches[best_match_index]:name = self.known_face_names[best_match_index]# 在帧上绘制矩形和标签cv2.rectangle(frame, (left, top), (right, bottom), (0, 0, 255), 2)cv2.rectangle(frame, (left, bottom - 35), (right, bottom), (0, 0, 255), cv2.FILLED)font = cv2.FONT_HERSHEY_DUPLEXcv2.putText(frame, name, (left + 6, bottom - 6), font, 1.0, (255, 255, 255), 1)# 显示结果cv2.imshow('Video', frame)# 按q键退出if cv2.waitKey(1) & 0xFF == ord('q'):break# 释放摄像头video_capture.release()cv2.destroyAllWindows()

我们在 real_time_face_recognition 方法中增加了 camera_index 参数，以便可以选择不同的摄像头。同时，我们使用了 tolerance 和 model 参数来调整人脸匹配的阈值和检测模型。

7. 总结

通过上述步骤，我们成功实现了基于Python的人脸识别算法，并将其封装为一个类库。使用OpenCV和dlib库，我们可以轻松地加载和编码已知人脸，并在实时视频流中检测和识别人脸。希望这篇文章对你有所帮助！