Grouping images based on the face in the images

Question

I have roughly 5000 photographs of various persons that I am organising according to the person. As a result, I intend to develop a Python script to carry out the task. I'm new to the Opencv library and am unsure whether this is feasible to do. To locate the faces in the photos, I wrote the code below.

import cv2

def find_faces(imagePath,results_path):

faceCascade = cv2.CascadeClassifier(cv2.data.haarcascades + "haarcascade_frontalface_default.xml")
eye_cascade = cv2.CascadeClassifier(cv2.data.haarcascades + "haarcascade_eye.xml")

image = cv2.imread(imagePath)
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)

faces = faceCascade.detectMultiScale(gray,scaleFactor=1.3,minNeighbors=3,minSize=(30, 30))

for (x, y, w, h) in faces:
    cv2.rectangle(image, (x, y), (x + w, y + h), (0, 255, 0), 2)
    roi_color = image[y:y + h, x:x + w]
    roi_gray = gray[y:y+h, x:x+w]
    eyes = eye_cascade.detectMultiScale(roi_gray)
    if len(eyes) >= 2:
        cv2.imwrite(results_path + str(w) + str(h) + '_faces.jpg', roi_color)

Now I need to organise the photographs per person.Any help with the next steps is greatly appreciated.

user_15171133 · Answer

As a fan of dlib / face_recognition libs I'd propose to use them (look at face_recognition tutorial - it's really easy)) to:

Create a dictionary with face encodings for all images and pathes to the image files:

Loop through the face encodings and compare each one with the rest: you don't need 100% similarity, normally 72..75% (0.28..0.25) means the same individual. I'm using something like that to store face data:

 class Face_Dictionary():
     """[The class contains all data structures to store in memory as 
        well as load and save to file
        face encodings data, including picture file full path and 
        face location for each face on every picture]
     """    

 def __init__(self, dicfilename="", mode="load"): #Create all data objects
     self.mode = mode
     self.dicfilename = dicfilename
     self.fl_Loaded = False
     self.fl_Saved = False
     self.Encodings = []
     self.Names = []
     self.facelocs = []
     self.fd = {"encodings": self.Encodings, "names": self.Names, 
                 "locations": self.facelocs}
     if bool(self.dicfilename) and self.mode == "load":
         self.fl_Loaded = self.load()


 def __del__(self): #redefine method del()
     del(self.Encodings)
     del(self.Names)
     del(self.facelocs)
     del(self.fd)

 def load(self):
     """
     [Loads a dictionary with face encodings from Pickle-type file into self.fd]

     Args:
         dicfilename ([str]): [Pickle-type file *.pkl]

     Returns:
         [bool]: [True if loaded. Also sets self.fl_Loaded as True]
     """
     if bool(self.dicfilename) and self.mode == "load":
         try:
             f = open(self.dicfilename, "rb")
         except (IOError, EOFError) as e:
             print("[ERR] Не можу знайти/прочитати файл кодувань обличь: %s" % e)
             return False
         else:
             if os.path.getsize(self.dicfilename) > 15:
                 self.fd = load(f)
                 self.fl_Loaded = True
             else:
                 f.close()
                 return  False
         f.close()
         return True
     else:
         return False

 def save(self):
     """
     [Saves dictionary with face encodings to Pickle-type file]

     Args:
         None

     Returns:
         [bool]: [True if self.fd has been saved to the original .
         Also sets self.fl_Saved as True]
     """
     if bool(self.dicfilename) and self.mode == "save":
         try:
             f = open(self.dicfilename, "wb")
         except OSError:
             print("[ERR] Не можу створити файл бази даних %s" % self.dicfilename)
             return False
         dump(self.fd, f, protocol=HIGHEST_PROTOCOL)
         f.close()
         self.fl_Saved = True
         return True

 def save_as(self, filename):
     """
     [Saves dictionary with face encodings to Pickle-type file with specified name]

     Args:
         filename ([str]): [Pickle-type file *.pkl]

     Returns:
         [bool]: [True if self.fd has been saved to the original .
         Also sets self.fl_Saved as True]
     """
     if bool(filename):
         try:
             f = open(filename, "wb")
         except OSError:
             print("[ERR] Не можу створити файл бази даних %s" % filename)
             return False
         dump(self.fd, f, protocol=HIGHEST_PROTOCOL)
         f.close()
         self.fl_Saved = True
         return True

Grouping images based on the face in the images

1 Answers