中商华兴建设有限公司网站深圳百度代理

1.人脸识别准备

使用的两个opencv包

D:\python2023>pip list |findstr opencv
opencv-contrib-python     4.8.1.78
opencv-python             4.8.1.78

数据集使用前一篇Javacv的数据集,网上随便找的60张图片,只是都挪到了D:\face目录下方便遍历

D:\face\1 30张刘德华图片
D:\face\2 30张刘亦菲图片

2.人脸识别模型训练

# -*- coding: utf-8 -*-
import osimport cv2
import numpy as nprecognizer = cv2.face.LBPHFaceRecognizer().create() # Fisher需要reshape
classifier = cv2.CascadeClassifier('E:\opencv\sources\data\haarcascades\haarcascade_frontalface_default.xml')
def load_dataset(dataset_path):images=[]labels=[]for root,dirs,files in os.walk(dataset_path):for file in files:images.append(cv2.imread(os.path.join(root, file),cv2.IMREAD_GRAYSCALE))labels.append(int(os.path.basename(root)))return images,labels
if __name__ == '__main__':images,labels = load_dataset('D:\\face')recognizer.train(images,np.array(labels))recognizer.save('face_model.xml')

3.人脸识别推理预测

# -*- coding: utf-8 -*-
import osimport cv2def face_detect(image):gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)classifier = cv2.CascadeClassifier('E:\opencv\sources\data\haarcascades\haarcascade_frontalface_default.xml')faces = classifier.detectMultiScale(gray, 1.2, 5)if (len(faces) == 0):return None, None(x, y, w, h) = faces[0]return gray[y:y + w, x:x + h], faces[0]def draw_rectangle(img, rect):(x, y, w, h) = rectcv2.rectangle(img, (x, y), (x + w, y + h), (255, 255, 0), 2)def draw_text(img, text, x, y):cv2.putText(img, text, (x, y), cv2.FONT_HERSHEY_COMPLEX, 1, (128, 128, 0), 2)def predict(image):image_copy = image.copy()face, rect = face_detect(image_copy)tuple = recognizer.predict(face)print(tuple)draw_rectangle(image_copy, rect)draw_text(image_copy, str(tuple[0]), rect[0], rect[1])return image_copyif __name__ == '__main__':recognizer = cv2.face.LBPHFaceRecognizer().create()  # Fisher需要reshaperecognizer.read("face_model.xml")for root, dirs, files in os.walk('D:\\face\\2'):for file in files:file_path = os.path.join(root, file)predict_image = predict(cv2.imread(file_path))cv2.imshow('result', predict_image)cv2.waitKey(1000)

总结

代码逻辑基本同Javacv,但更简洁，这里训练出来模型准确度也高于Javacv (可能是参数不一致导致的)