如何修改Python KNN人脸识别代码，识别未注册人脸显示Unknown

原代码无法识别未注册人脸的核心问题是：没有设置置信度阈值，且判断逻辑无效（len(LABELS) >= 0永远成立）。以下是修改后的完整代码及关键改动说明：

修改后的代码 #

from sklearn.neighbors import KNeighborsClassifier
import cv2
import pickle
import numpy as np
import os
import csv
import time
from datetime import datetime


video = cv2.VideoCapture(0)
facedetect = cv2.CascadeClassifier('data/haarcascade_frontalface_default.xml')

with open('data/names.pkl', 'rb') as w:
    LABELS = pickle.load(w)
with open('data/faces_data.pkl', 'rb') as f:
    FACES = pickle.load(f)

print('Shape of Faces matrix --> ', FACES.shape)

# 初始化KNN分类器
knn = KNeighborsClassifier(n_neighbors=5)
knn.fit(FACES, LABELS)

imgBackground = None

COL_NAMES = ['NAME', 'TIME']
# 设置置信度阈值，可根据实际场景调整
CONFIDENCE_THRESHOLD = 0.6

while True:
    ret, frame = video.read()
    gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
    faces = facedetect.detectMultiScale(gray, 1.3, 5)
    for (x, y, w, h) in faces:
        crop_img = frame[y:y+h, x:x+w, :]
        resized_img = cv2.resize(crop_img, (50,50)).flatten().reshape(1,-1)
        
        # 获取预测结果及对应概率
        output = knn.predict(resized_img)
        prob_scores = knn.predict_proba(resized_img)
        max_prob = np.max(prob_scores)
        
        ts = time.time()
        date = datetime.fromtimestamp(ts).strftime("%d-%m-%Y")
        timestamp = datetime.fromtimestamp(ts).strftime("%H:%M:%S")
        exist = os.path.isfile("Attendance/Attendance_" + date + ".csv")
        
        # 基于概率判断是否为未知人脸
        if max_prob >= CONFIDENCE_THRESHOLD:
            recognized_name = output[0]
        else:
            recognized_name = "Unknown"

        attendance = [recognized_name, str(timestamp)]
        
        # 绘制人脸框和识别结果
        cv2.rectangle(frame, (x,y), (x+w, y+h), (0,0,255), 1)
        cv2.rectangle(frame,(x,y),(x+w,y+h),(50,50,255),2)
        cv2.rectangle(frame,(x,y-40),(x+w,y),(50,50,255),-1)
        cv2.putText(frame, recognized_name, (x,y-15), cv2.FONT_HERSHEY_COMPLEX, 1, (255,255,255), 1)
        cv2.rectangle(frame, (x,y), (x+w, y+h), (50,50,255), 1)
        
    if imgBackground is not None and not imgBackground.empty():
        imgBackground[162:162 + frame.shape[0], 55:55 + frame.shape[1]] = frame
        cv2.imshow("Frame", imgBackground)
    else:
        cv2.imshow("Frame", frame)

    k = cv2.waitKey(1)
    if k == ord('o'):
        if exist:
            with open("Attendance/Attendance_" + date + ".csv", "+a") as csvfile:
                writer = csv.writer(csvfile)
                writer.writerow(attendance)
        else:
            with open("Attendance/Attendance_" + date + ".csv", "+a") as csvfile:
                writer = csv.writer(csvfile)
                writer.writerow(COL_NAMES)
                writer.writerow(attendance)
    if k == ord('q'):
        break
video.release()
cv2.destroyAllWindows()

关键改动说明 #

• 添加CONFIDENCE_THRESHOLD置信度阈值：通过predict_proba获取每个类别的预测概率，取最高概率与阈值对比，低于阈值则判定为未知人脸。
• 替换无效判断逻辑：移除len(LABELS) >=0的无效判断，改为基于概率的有效判定逻辑。
• 更新显示文本：将cv2.putText的显示内容从output[0]改为recognized_name，确保界面显示最终判定结果。
• 阈值可调：可根据实际场景调整CONFIDENCE_THRESHOLD数值——误判较多时提高阈值，漏判较多时降低阈值。

内容的提问来源于stack exchange，提问作者Reungrit Saejiw