Face_recognition: Unable to identify matched image

[shehzi-khan]

• face_recognition version:
• Python version: 2.7
• Operating System: Linux Mint (Sarah)

Description

I’m trying to create an application to recognize any of the known faces from a video stream. I have a directory “training_images” which contains one image for each known face. I’m following the example code given by you and slightly modified it to add multiple known faces.
In the response of face_recognition .compare_faces(), I was expecting a list of matched results, containing a set of 128 boolean values for each training image. Instead of that it returns only one set of 128 boolean values. While which training_image_encoding was matched is still unknown.

Can you help me, where I’m doing wrong?

What I Did

import face_recognition
import cv2
import numpy
import os

video_capture = cv2.VideoCapture(0)


files= os.listdir("training_images")
print files
training_face_encoding=[]
for file in files:
    print file
    training_image = face_recognition.load_image_file('training_images/'+file)
    training_face_encoding.append(face_recognition.face_encodings(training_image)[0])

face_locations = []
face_encodings = []
face_names = []
process_this_frame = True
i =0
while True:
    ret, frame = video_capture.read()
    frame=cv2.flip(frame,1)
    small_frame = cv2.resize(frame, (0, 0), fx=0.25, fy=0.25)

    if i < 4:
        process_this_frame=False
        i +=1
    else:
        i=0
    if process_this_frame:
        face_locations = face_recognition.face_locations(small_frame)
        face_encodings = face_recognition.face_encodings(small_frame, face_locations)

        face_names = []
        for face_encoding in face_encodings:
            match = face_recognition.compare_faces([training_face_encoding], face_encoding)
            name = "Unknown"
            print match[0]
            print files

            i =0
            for m in match[0]:
                # print m
                # print i
                if m.all():
                    name = files[i]
                    face_names.append(name)
                i +=1




    process_this_frame = not process_this_frame


    for (top, right, bottom, left), name in zip(face_locations, face_names):
        top *= 4
        right *= 4
        bottom *= 4
        left *= 4

        cv2.rectangle(frame, (left, top), (right, bottom), (0, 0, 255), 2)

        cv2.rectangle(frame, (left, bottom - 35), (right, bottom), (0, 0, 255), cv2.FILLED)
        font = cv2.FONT_HERSHEY_DUPLEX
        cv2.putText(frame, name, (left + 6, bottom - 6), font, 1.0, (255, 255, 255), 1)

    cv2.imshow('Video', frame)

    if cv2.waitKey(1) & 0xFF == ord('q'):
        break

video_capture.release()
cv2.destroyAllWindows()

[shehzi-khan]

The application should read all training images from "training_images" directory, then match each frame of video stream to these images and in the end tell the name of most appropriately matched image file.

[shehzi-khan]

It will be more suitable if I can add multiple images of same person and application tells the name of the person in video frame.

[hellopatrick]

Instead of that it returns only one set of 128 boolean values.

Isn't this what it should be? Each boolean value, match[i], says if training_face_encoding[i] is within the threshold (default=0.6) of face_encoding.

face_recognition.compare_faces([training_face_encoding], face_encoding)

This should just be face_recognition.compare_faces(training_face_encoding, face_encoding). If you were just copying the example, the obama_face_encoding is a lone encoding and needed to be placed in a list for the function, you don't have to do that.

With that change you should just be able to iterate over match looking for the True values.

[ageitgey]

Here's a fixed version of your code. However I only tested it in Python 3 (not Python 2):

import face_recognition
import cv2
import numpy
import os

video_capture = cv2.VideoCapture(0)

files = os.listdir("training_images")
print(files)
training_face_encoding=[]
for file in files:
    print(file)
    training_image = face_recognition.load_image_file('training_images/'+file)
    training_face_encoding.append(face_recognition.face_encodings(training_image)[0])

face_locations = []
face_encodings = []
face_names = []
process_this_frame = True
i = 0
while True:
    ret, frame = video_capture.read()
    frame=cv2.flip(frame, 1)
    small_frame = cv2.resize(frame, (0, 0), fx=0.25, fy=0.25)

    if i < 4:
        process_this_frame=False
        i += 1
    else:
        i = 0
    if process_this_frame:
        face_locations = face_recognition.face_locations(small_frame)
        face_encodings = face_recognition.face_encodings(small_frame, face_locations)

        face_names = []
        for face_encoding in face_encodings:
            match = face_recognition.compare_faces(training_face_encoding, face_encoding)
            name = "Unknown"
            print(match)
            print(files)

            i = 0
            for m in match:
                if m:
                    name = files[i]
                i += 1

            face_names.append(name)

    process_this_frame = not process_this_frame

    for (top, right, bottom, left), name in zip(face_locations, face_names):
        top *= 4
        right *= 4
        bottom *= 4
        left *= 4

        cv2.rectangle(frame, (left, top), (right, bottom), (0, 0, 255), 2)

        cv2.rectangle(frame, (left, bottom - 35), (right, bottom), (0, 0, 255), cv2.FILLED)
        font = cv2.FONT_HERSHEY_DUPLEX
        cv2.putText(frame, name, (left + 6, bottom - 6), font, 1.0, (255, 255, 255), 1)

    cv2.imshow('Video', frame)

    if cv2.waitKey(1) & 0xFF == ord('q'):
        break

video_capture.release()
cv2.destroyAllWindows()

[ageitgey]

Output:

$ python3 test.py
['Adam.jpg', 'biden.jpg', 'obama.jpg', 'obama2.jpg', 'obama_small.jpg']
Adam.jpg
biden.jpg
obama.jpg
obama2.jpg
obama_small.jpg
[True, False, False, False, False]
['Adam.jpg', 'biden.jpg', 'obama.jpg', 'obama2.jpg', 'obama_small.jpg']

The biggest issue was that you are were accidentally passing in an array of an array to face_recognition.compare_faces() instead of just passing in a normal array like @hellopatrick said.

[31893]

Thank you for this wonderful model.
I am using

• face_recognition version:

• Python version: 2.7.5

• Operating System: ubuntu(16.04)
  I had collected few faces and created my dataset .
  Does the accuracy of recognizing people reduce if we increase the size of the dataset by any chance?

[ageitgey]

@31893 No, the accuracy isn't based on the number of photos you are testing against (if I understand your question correctly)