Real time Face Recognition Blog

face-expression-recognition.py

# capture the video from default camera
# video_stream = cv2.VideoCapture(0)

# read video from video file
video_file_path = 'test/face-demographics-walking.mp4'
video_stream = cv2.VideoCapture(video_file_path)

# initialize the number of frame needed to be skipped
skip = 0

while True:

  #Read the frame
  for i in range(skip):
      video_stream.grab()

  # get current frame
  stat, frame = video_stream.read()

  # check that frame is left to read
  if stat == False:
    print('Video has ended')
    break  

  # reduce size of image to 50%
  scaled_frame = cv2.resize(frame, (0,0), fx=0.5, fy=0.5)

  # find all face locations using face_locations() function
  # model can be "cnn" or "hog"
  # number_of_times_to_upsample = 1 higher and detect more faces

  all_face_locations = face_recognition.face_locations(scaled_frame, model="cnn")

  for index,face_location in enumerate(all_face_locations):

    # split the tuple 
    top_pos, right_pos, bottom_pos, left_pos = face_location

    # scale up the coordinates by 2 times
    top_pos = top_pos * 2
    right_pos = right_pos * 2
    bottom_pos = bottom_pos * 2
    left_pos = left_pos * 2

    #---------------------------------------------------------------------------
    # Face Expression Detection
    #---------------------------------------------------------------------------

    # Slice frame image array by positions face_locations
    face_image = frame[top_pos:bottom_pos, left_pos:right_pos]

    # preprocessing input, convert it to an image like as the data in dataset
    # convert face image to grayscale
    face_image = cv2.cvtColor(face_image, cv2.COLOR_BGR2GRAY)

    # resize to 48X48 px size
    face_image = cv2.resize(face_image, (48, 48))

    # convert the PIL image into a 3D numpy array
    img_pixels = image.img_to_array(face_image)

    # expand the shape of an array into single row multiple columns
    img_pixels = np.expand_dims(img_pixels, axis = 0)

    # pixels are in range of [0, 255]. Normalize all pixels in scale of [0, 1]
    img_pixels /= 255

    # do prediction using model, get the prediction values for all 7 expressions
    exp_predictions = face_exp_model.predict(img_pixels)

    # find max indexed prediction value (0 till 7)
    max_index = np.argmax(exp_predictions[0])

    # get the corresponding label from emotions_label list
    emotion_label = emotions_label[max_index]
    
    # draw rectangle around each face
    cv2.rectangle(frame, (left_pos,top_pos), (right_pos,bottom_pos), (0,0,255), 2)

    # disply the name of emotion as text along the face
    font = cv2.FONT_HERSHEY_SIMPLEX
    cv2.rectangle(frame, (left_pos-1, bottom_pos + 18), (right_pos+1, bottom_pos), (0, 0, 255), -1)
    cv2.putText(frame, emotion_label, (left_pos+2,bottom_pos+11), font, 0.5, (255,255,255), 1)
    
    #---------------------------------------------------------------------------

  # display frame
  cv2.imshow("Video",frame)

  # for only google colab
  # cv2_imshow(frame)

  # Press 'q' on the keyboard to break the while loop
  if cv2.waitKey(1) & 0xFF == ord('q'):
    break
    
# Release the video stream resource
video_stream.release()
cv2.destroyAllWindows()

import-libraries.py

# importing required libraries
import cv2
import numpy as np
from keras.preprocessing import image
from keras.models import model_from_json
import  face_recognition

# only for google colab
# from google.colab.patches import cv2_imshow

initialize-model-and-weights.py

# face expression model initialization
face_exp_model = model_from_json(open("models/facial_expression_model_structure.json","r").read())

# load weights into model
face_exp_model.load_weights("models/facial_expression_model_weights.h5")

# list of emotions labels
emotions_label = ('Angry','Digust','Fear','Happy','Sad','Surprise','Neutral')

realtime-face-expression-recognition.py

# importing required libraries
import cv2
import numpy as np
from keras.preprocessing import image
from keras.models import model_from_json
import  face_recognition

# only for google colab
# from google.colab.patches import cv2_imshow

# face expression model initialization
face_exp_model = model_from_json(open("models/facial_expression_model_structure.json","r").read())

# load weights into model
face_exp_model.load_weights("models/facial_expression_model_weights.h5")

# list of emotions labels
emotions_label = ('Angry','Digust','Fear','Happy','Sad','Surprise','Neutral')

# capture the video from default camera
# video_stream = cv2.VideoCapture(0)

# read video from video file
video_file_path = 'test/face-demographics-walking.mp4'
video_stream = cv2.VideoCapture(video_file_path)

# initialize the number of frame needed to be skipped
skip = 0

while True:

  #Read the frame
  for i in range(skip):
      video_stream.grab()

  # get current frame
  stat, frame = video_stream.read()

  # check that frame is left to read
  if stat == False:
    print('Video has ended')
    break  

  # reduce size of image to 50%
  scaled_frame = cv2.resize(frame, (0,0), fx=0.5, fy=0.5)

  # find all face locations using face_locations() function
  # model can be "cnn" or "hog"
  # number_of_times_to_upsample = 1 higher and detect more faces

  all_face_locations = face_recognition.face_locations(scaled_frame, model="cnn")

  for index,face_location in enumerate(all_face_locations):

    # split the tuple 
    top_pos, right_pos, bottom_pos, left_pos = face_location

    # scale up the coordinates by 2 times
    top_pos = top_pos * 2
    right_pos = right_pos * 2
    bottom_pos = bottom_pos * 2
    left_pos = left_pos * 2

    #---------------------------------------------------------------------------
    # Face Expression Detection
    #---------------------------------------------------------------------------

    # Slice frame image array by positions face_locations
    face_image = frame[top_pos:bottom_pos, left_pos:right_pos]

    # preprocessing input, convert it to an image like as the data in dataset
    # convert face image to grayscale
    face_image = cv2.cvtColor(face_image, cv2.COLOR_BGR2GRAY)

    # resize to 48X48 px size
    face_image = cv2.resize(face_image, (48, 48))

    # convert the PIL image into a 3D numpy array
    img_pixels = image.img_to_array(face_image)

    # expand the shape of an array into single row multiple columns
    img_pixels = np.expand_dims(img_pixels, axis = 0)

    # pixels are in range of [0, 255]. Normalize all pixels in scale of [0, 1]
    img_pixels /= 255

    # do prediction using model, get the prediction values for all 7 expressions
    exp_predictions = face_exp_model.predict(img_pixels)

    # find max indexed prediction value (0 till 7)
    max_index = np.argmax(exp_predictions[0])

    # get the corresponding label from emotions_label list
    emotion_label = emotions_label[max_index]
    
    # draw rectangle around each face
    cv2.rectangle(frame, (left_pos,top_pos), (right_pos,bottom_pos), (0,0,255), 2)

    # disply the name of emotion as text along the face
    font = cv2.FONT_HERSHEY_SIMPLEX
    cv2.rectangle(frame, (left_pos-1, bottom_pos + 18), (right_pos+1, bottom_pos), (0, 0, 255), -1)
    cv2.putText(frame, emotion_label, (left_pos+2,bottom_pos+11), font, 0.5, (255,255,255), 1)
    
    #---------------------------------------------------------------------------

  # display frame
  cv2.imshow("Face Expressions Video",frame)

  # for only google colab
  # cv2_imshow(frame)

  # Press 'q' on the keyboard to break the while loop
  if cv2.waitKey(1) & 0xFF == ord('q'):
    break
    
# Release the video stream resource
video_stream.release()
cv2.destroyAllWindows()