[PYTHON] Face recognition / cutting with OpenCV

Introduction

I modified the code that recognizes the face in the image and encloses it in a square frame, and created a code that cuts out as a file.

Library import

Import libraries such as OpenCV and Pillow.

import os
import subprocess
from PIL import Image
import cv2 as cv

dir0 = 'org' 
dir1 = 'png'
dir2 = 'png_resize'
dir3 = 'face'

Image preparation

Place the original image (jpg, size 500x334px) in the org folder. Convert this to a png file, adjust the size width to 600px, and store it in the png_resize folder.

files0 = os.listdir(dir0)
files0.sort()

for file in files0:

    if '.jpg'  in file:        
        command = 'sips --setProperty format png ' + dir0 +'/' + file +  ' --out ' + dir1 +'/' +  file.replace('.jpg','.png') 
        subprocess.call(command, shell=True)
        print(file) 

files1 = os.listdir(dir1)
files1.sort()

# aaa.jpg  

for file in files1:   
    if '.png' in file:   
        img0 = os.path.join(dir1, file)
        img0_img = Image.open(img0)
        h = img0_img.height
        w = img0_img.width
        img1_img = img0_img.resize((600,round(600*h/w)))
        img1 = os.path.join(dir2, file) 
        img1_img.save(img1)
        print(file) 

# aaa.png

files2 = os.listdir(dir2)
files2.sort()    

Face recognizer installation

Face classifiers are available at here.

face_cascade = cv.CascadeClassifier('haarcascade_frontalface_default.xml')
#face_cascade = cv.CascadeClassifier('haarcascade_profileface.xml')

Face recognizer execution, face image cutout

Cut out the face image. The cutout range is 10px wider up, down, left and right than the detection range.

for file in files2:
    if '.png' in file:
        dirfile = os.path.join(dir2, file) 
        img = cv.imread(dirfile)
        gray = cv.cvtColor(img, cv.COLOR_BGR2GRAY)
        
        faces = face_cascade.detectMultiScale(gray, 1.3, 5)
        
        for (x,y,w,h) in faces:
            face = img[y-10:y+h+10, x-10:x+w+10]
            face_name = str(file.strip('.png'))+'_'+str(x)+'_'+str(y)+'.png'
            dirface = os.path.join(dir3,face_name)
            facefile = cv.imwrite(dirface, face) 
            #cv.rectangle(img,(x-10,y-10),(x+w+10,y+h+10),(255,0,0),2)
            print(face_name)  

# aaa_152_22.png
# aaa_11_70.png
# aaa_438_41.png
# aaa_79_106.png
# aaa_385_140.png
# aaa_190_175.png
# aaa_269_171.png
# aaa_76_206.png
# aaa_527_257.png
# aaa_91_277.png
# aaa_254_330.png
# aaa_446_348.png  

A total of 12 face images were cut out from the above images. If you can see most of the face, you know that the cutout was successful.

At the end

It seems that it can be used when making a list with a photo from a group photo. It would be convenient for anyone to use the trained classifier if it was published as a cascade file.

reference

https://docs.opencv.org/4.1.0/d7/d8b/tutorial_py_face_detection.html https://github.com/opencv/opencv/tree/master/data/haarcascades