In this post, I am going to explain how to recognize faces using OpenCV built-in Face recognizer.

 

You can download this example from below link.

https://github.com/harikrishna553/python/tree/main/opencv/1.images/face_recognition

 

 

Train the model with images and respective labels

I am using haar cascade face recognizer to detect the faces in given images.

 

haar_cascade = cv.CascadeClassifier('haarcascade_frontalface_default.xml')

 

Project hierarchy looks like below.

 

I collected some images of Mahendra singh Dhoni, Virender Sehwag and kept them in ‘Dhoni’ and ‘Sehwag’ folder.

 

We need to loop through the images one by one, extract the face coordinates and map the labels to them.

 

Following snippet do the above one.

peoples = ['Dhoni', 'Sehwag']

features = []
labels = []

def populate_face_rect_and_label():
    for person in peoples:
        path = os.path.join('.', person)
        label = peoples.index(person)

        for image in os.listdir(path):
            img_path = os.path.join(path, image)

            image_array = cv.imread(img_path)
            gray_scaled_image = cv.cvtColor(image_array, cv.COLOR_BGR2GRAY)
            faces_rectangle = haar_cascade.detectMultiScale(gray_scaled_image, scaleFactor=1.1, minNeighbors=7)
            for (a, b, c, d) in faces_rectangle:
                face_array = gray_scaled_image[b:b+d, a:a+c]
                features.append(face_array)
                labels.append(label)

Once face coordinates and their respective labels are identified, we need to train the model.

face_recognizer = cv.face_LBPHFaceRecognizer.create()  # Use create() to create the recognizer

features = np.array(features, dtype='object')
labels = np.array(labels, dtype='int')  # Use 'int' instead of 'uint8'

face_recognizer.train(features, labels)

Test the image with trained model

Extract the face coordinates from trained model.

image_to_test = cv.imread('img_to_test.png')
gray_scale_image_to_test = cv.cvtColor(image_to_test, cv.COLOR_BGR2GRAY)
face_rect = haar_cascade.detectMultiScale(gray_scale_image_to_test, scaleFactor=1.3, minNeighbors=5)

For each recognized face in the test image, extract the face coordinates, and predict the label using face recognizer predict method

for (a, b, c, d) in face_rect:
    face = gray_scale_image_to_test[b:b+d, a:a+c]
    label, confidence = face_recognizer.predict(face)
    print(f'Label {peoples[label]} with Confidence {confidence}')
    cv.putText(image_to_test, str(f'Label {peoples[label]} with confidence {confidence}'), (50, 50), cv.FONT_HERSHEY_SIMPLEX, 1.0, (255, 0, 255), thickness=4)
    cv.rectangle(image_to_test, (a, b), (a+c, b+d), color=(0, 255, 0), thickness=3)

Find the below working application.

 

face_recognizer.py

import os
import cv2 as cv
import numpy as np

haar_cascade = cv.CascadeClassifier('haarcascade_frontalface_default.xml')

peoples = ['Dhoni', 'Sehwag']

features = []
labels = []

def populate_face_rect_and_label():
    for person in peoples:
        path = os.path.join('.', person)
        label = peoples.index(person)

        for image in os.listdir(path):
            img_path = os.path.join(path, image)

            image_array = cv.imread(img_path)
            gray_scaled_image = cv.cvtColor(image_array, cv.COLOR_BGR2GRAY)
            faces_rectangle = haar_cascade.detectMultiScale(gray_scaled_image, scaleFactor=1.1, minNeighbors=7)
            for (a, b, c, d) in faces_rectangle:
                face_array = gray_scaled_image[b:b+d, a:a+c]
                features.append(face_array)
                labels.append(label)

populate_face_rect_and_label()

face_recognizer = cv.face_LBPHFaceRecognizer.create()  # Use create() to create the recognizer

features = np.array(features, dtype='object')
labels = np.array(labels, dtype='int')  # Use 'int' instead of 'uint8'

face_recognizer.train(features, labels)
# face_recognizer.save('trained_face_recognizer.yml')  # Save the trained model

image_to_test = cv.imread('img_to_test.png')
gray_scale_image_to_test = cv.cvtColor(image_to_test, cv.COLOR_BGR2GRAY)
face_rect = haar_cascade.detectMultiScale(gray_scale_image_to_test, scaleFactor=1.3, minNeighbors=5)

for (a, b, c, d) in face_rect:
    face = gray_scale_image_to_test[b:b+d, a:a+c]
    label, confidence = face_recognizer.predict(face)
    print(f'Label {peoples[label]} with confidence {confidence}')
    cv.putText(image_to_test, str(f'Label {peoples[label]} with confidence {confidence}'), (50, 50), cv.FONT_HERSHEY_SIMPLEX, 1.0, (255, 0, 255), thickness=4)
    cv.rectangle(image_to_test, (a, b), (a+c, b+d), color=(0, 255, 0), thickness=3)

cv.imshow('Face Recognition', image_to_test)
cv.waitKey(0)
cv.destroyAllWindows()

#print(f'Features length: {len(features)}')
#print(f'Labels length: {len(labels)}')

Previous                                                    Next                                                    Home