def prediction_path(path):
# load keras model
model = define_model()
model = model_weights(model)
# list of given emotions
EMOTIONS = [
'Angry', 'Disgusted', 'Fearful', 'Happy', 'Sad', 'Surprised', 'Neutral'
]
if os.path.exists(path):
# read the image
img = cv2.imread(path, 0)
# check if image is valid or not
if img is None:
print('Invalid image !!')
return
else:
print('Image not found')
return
# resize image for the model
img = cv2.resize(img, (48, 48))
img = np.reshape(img, (1, 48, 48, 1))
# do prediction
result = model.predict(img)
print('Detected emotion: ' + str(EMOTIONS[np.argmax(result[0])]))
return
def prepare_realtime_emotions():
global model,faceCascade,emoji_faces
# load keras model
model = define_model()
model = model_weights(model)
print('Model loaded')
# load haar cascade for face
faceCascade = cv2.CascadeClassifier(r'haarcascades/haarcascade_frontalface_default.xml')
# list of given emotions
# store the emoji coreesponding to different emotions
emoji_faces = []
for index, emotion in enumerate(EMOTIONS):
emoji_faces.append(cv2.imread('emojis/' + emotion.lower() + '.png', -1))
def prediction_path(img_str):
# load keras model
cv2.ocl.setUseOpenCL(False)
model = model_weights(define_model())
path = './logs/' + time.strftime("%Y%m%d-%H%M%S") + '.png'
with open(path, "wb") as fh:
fh.write(base64.b64decode(img_str))
logger.info('Image save at ' + path)
img = cv2.imread(path)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
img_rot = cv2.rotate(img, cv2.ROTATE_180)
gray_rot = cv2.cvtColor(img_rot, cv2.COLOR_BGR2GRAY)
face_cascade = cv2.CascadeClassifier('./haarcascade_frontalface_alt2.xml')
# Detect faces
faces = face_cascade.detectMultiScale(
gray,
scaleFactor=1.2,
minNeighbors=5,
minSize=(20, 20),
)
faces_rot = face_cascade.detectMultiScale(
gray_rot,
scaleFactor=1.2,
minNeighbors=5,
minSize=(20, 20),
)
if (len(faces) >= 1):
logger.info('Image not rotated')
return do_prediction(faces, img, model)
elif (len(faces_rot) >= 1):
logger.info('Image rotated')
return do_prediction(faces_rot, img_rot, model)
else:
return cs.SPECIAL_EMOTIONS[1]
def prediction_path(path):
model = define_model()
model = model_weights(model)
images = cv2.imread(path)
gray_img = cv2.cvtColor(images, cv2.COLOR_BGR2GRAY)
face_haar_cascade = cv2.CascadeClassifier(
'haarcascade_frontalface_default.xml')
faces_detected = face_haar_cascade.detectMultiScale(
gray_img,
scaleFactor=1.1,
minNeighbors=5,
minSize=(30, 30),
flags=cv2.CASCADE_SCALE_IMAGE)
print(faces_detected)
print(type(faces_detected))
for (x, y, w, h) in faces_detected:
cv2.rectangle(images, (x, y), (x + w, y + h), (255, 0, 0))
roi_gray = gray_img[
y:y + w,
x:x + h] #cropping region of interest i.e. face area from image
roi_gray = cv2.resize(roi_gray, (48, 48))
roi_gray = np.reshape(roi_gray, (1, 48, 48, 1))
mean = np.mean(roi_gray)
std = np.std(roi_gray)
roi_gray = (roi_gray - mean) / (std + 1e-7)
predictions = model.predict(roi_gray)
print(predictions)
#find max indexed array
max_index = np.argmax(predictions[0])
emotions = ('anger', 'happy', 'neutral', 'sad')
predicted_emotion = emotions[max_index]
cv2.putText(images, predicted_emotion, (int(x), int(y)),
cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 255), 2)
#resized_img = cv2.resize(images, (int(images.shape[0]/2), int(images.shape[1]/2)))
cv2.imshow("result", images)
cv2.waitKey(0)
cv2.destroyAllWindows()
def main():
model = define_model()
model = model_weights(model)
print('Model loaded')
result = np.array((1,7))
faceCascade = cv2.CascadeClassifier(r'haarcascades/haarcascade_frontalface_default.xml')
# list of given emotions
EMOTIONS = ['Angry', 'Disgusted', 'Fearful', 'Happy', 'Sad', 'Surprised', 'Neutral']
# store the emoji coreesponding to different emotions
emoji_faces = []
save_loc = 'save_loc/1.jpg'
for index, emotion in enumerate(EMOTIONS):
emoji_faces.append(cv2.imread('emojis/' + emotion.lower() + '.png', -1))
# set video capture device , webcam in this case
video_capture = cv2.VideoCapture(0)
video_capture.set(3, 640) # WIDTH
video_capture.set(4, 480) # HEIGHT
prev_time = time.time()
once =False
while True:
ret, frame = video_capture.read()
# mirror the frame
frame = cv2.flip(frame, 1, 0)
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
# find face in the frame
faces = faceCascade.detectMultiScale(
gray,
scaleFactor=1.1,
minNeighbors=5,
minSize=(30, 30),
flags=cv2.CASCADE_SCALE_IMAGE
)
for (x, y, w, h) in faces:
cv2.rectangle(frame, (x-10, y-70),(x+w+20, y+h+40), (15, 175, 61), 4)
# roi_gray = gray[y:y+h, x:x+w]
color_img = frame[y-90:y+h+70, x-50:x+w+50]
curr_time = time.time()
cv2.imwrite(save_loc, color_img)
img = cv2.imread(save_loc, 0)
if img is not None:
if curr_time - prev_time >=1:
img = cv2.resize(img, (48, 48))
img = np.reshape(img, (1, 48, 48, 1))
# do prediction
result = model.predict(img)
prev_time = time.time()
total_sum = np.sum(result[0])
once=True
# select the emoji face with highest confidence
emoji_face = emoji_faces[np.argmax(result[0])]
if once==True:
for index, emotion in enumerate(EMOTIONS):
text = str(round(Decimal(result[0][index]/total_sum*100), 2) ) + "%"
# for drawing progress bar
cv2.rectangle(frame, (100, index * 20 + 10), (100 +int(result[0][index] * 100), (index + 1) * 20 + 4),
(255, 0, 0), -1)
# for putting emotion labels
cv2.putText(frame, emotion, (10, index * 20 + 20),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (7, 109, 16), 2)
# for putting percentage confidence
cv2.putText(frame, text, (105 + int(result[0][index] * 100), index * 20 + 20),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 0, 0), 1)
for c in range(0, 3):
# for doing overlay we need to assign weights to both foreground and background
foreground = emoji_face[:, :, c] * (emoji_face[:, :, 3] / 255.0)
background = frame[350:470, 10:130, c] * (1.0 - emoji_face[:, :, 3] / 255.0)
frame[350:470, 10:130, c] = foreground + background
break
# Display the resulting frame
cv2.imshow('Video', frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
# When everything is done, release the capture
video_capture.release()
cv2.destroyAllWindows()
def realtime_emotions():
model = define_model()
model = model_weights(model)
save_loc = 'save_loc/1.jpg'
result = np.array((1, 7))
once = False
faceCascade = cv2.CascadeClassifier(
r'haarcascades/haarcascade_frontalface_default.xml')
EMOTIONS = [
'Angry', 'Disgusted', 'Fearful', 'Happy', 'Sad', 'Surprised', 'Neutral'
]
emoji_faces = []
for index, emotion in enumerate(EMOTIONS):
emoji_faces.append(cv2.imread('emojis/' + emotion + '.png', -1))
video_capture = cv2.VideoCapture(0)
video_capture.set(3, 640)
video_capture.set(4, 480)
prev_time = time.time()
# start webcam feed
while True:
#
ret, frame = video_capture.read()
frame = cv2.flip(frame, 1, 0)
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
faces = faceCascade.detectMultiScale(gray,
scaleFactor=1.1,
minNeighbors=5,
minSize=(30, 30),
flags=cv2.CASCADE_SCALE_IMAGE)
# Draw a rectangle around the faces
for (x, y, w, h) in faces:
roi_color = frame[y - 90:y + h + 70, x - 50:x + w + 50]
cv2.imwrite(save_loc, roi_color)
cv2.rectangle(frame, (x - 10, y - 70), (x + w + 20, y + h + 40),
(15, 175, 61), 4)
curr_time = time.time()
if curr_time - prev_time >= 1:
img = cv2.imread(save_loc, 0)
if img is not None:
once = True
img = cv2.resize(img, (48, 48))
img = np.reshape(img, (1, 48, 48, 1))
result = model.predict(img)
print(EMOTIONS[np.argmax(result[0])])
prev_time = time.time()
if once == True:
total_sum = np.sum(result[0])
emoji_face = emoji_faces[np.argmax(result[0])]
for index, emotion in enumerate(EMOTIONS):
text = str(
round(Decimal(result[0][index] / total_sum * 100),
2)) + "%"
cv2.rectangle(frame, (100, index * 20 + 10),
(100 + int(result[0][index] * 100),
(index + 1) * 20 + 4), (255, 0, 0), -1)
cv2.putText(frame, emotion, (10, index * 20 + 20),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (173, 9, 136),
2)
cv2.putText(
frame, text,
(105 + int(result[0][index] * 100), index * 20 + 20),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 55, 125), 1)
for c in range(0, 3):
foreground = emoji_face[:, :,
c] * (emoji_face[:, :, 3] / 255.0)
background = frame[350:470, 10:130,
c] * (1.0 - emoji_face[:, :, 3] / 255.0)
frame[350:470, 10:130, c] = foreground + background
break
# Display the resulting frame
cv2.imshow('Video', frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
video_capture.release()
cv2.destroyAllWindows()
def realtime_emotions():
# load keras model
model = define_model()
model = model_weights(model)
print('Model loaded')
# save location for image
save_loc = 'save_loc/1.jpg'
# numpy matrix for stroing prediction
result = np.array((1, 7))
# for knowing whether prediction has started or not
once = False
# load haar cascade for face
faceCascade = cv2.CascadeClassifier(
r'haarcascades/haarcascade_frontalface_default.xml')
# list of given emotions
EMOTIONS = [
'Angry', 'Disgusted', 'Fearful', 'Happy', 'Sad', 'Surprised', 'Neutral'
]
# store the emoji coreesponding to different emotions
emoji_faces = []
for index, emotion in enumerate(EMOTIONS):
emoji_faces.append(cv2.imread('emojis/' + emotion.lower() + '.png',
-1))
# set video capture device , webcam in this case
video_capture = cv2.VideoCapture(0)
video_capture.set(3, 640) # WIDTH
video_capture.set(4, 480) # HEIGHT
# save current time
prev_time = time.time()
# start webcam feed
while True:
# Capture frame-by-frame
ret, frame = video_capture.read()
# mirror the frame
frame = cv2.flip(frame, 1, 0)
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
# find face in the frame
faces = faceCascade.detectMultiScale(gray,
scaleFactor=1.1,
minNeighbors=5,
minSize=(30, 30),
flags=cv2.CASCADE_SCALE_IMAGE)
# Draw a rectangle around the faces
for (x, y, w, h) in faces:
# required region for the face
roi_color = frame[y - 90:y + h + 70, x - 50:x + w + 50]
# save the detected face
cv2.imwrite(save_loc, roi_color)
# draw a rectangle bounding the face
cv2.rectangle(frame, (x - 10, y - 70), (x + w + 20, y + h + 40),
(15, 175, 61), 4)
# keeps track of waiting time for emotion recognition
curr_time = time.time()
# do prediction only when the required elapsed time has passed
if curr_time - prev_time >= 1:
# read the saved image
img = cv2.imread(save_loc, 0)
if img is not None:
# indicates that prediction has been done atleast once
once = True
# resize image for the model
img = cv2.resize(img, (48, 48))
img = np.reshape(img, (1, 48, 48, 1))
# do prediction
result = model.predict(img)
print(EMOTIONS[np.argmax(result[0])])
#save the time when the last face recognition task was done
prev_time = time.time()
if once == True:
total_sum = np.sum(result[0])
# select the emoji face with highest confidence
emoji_face = emoji_faces[np.argmax(result[0])]
for index, emotion in enumerate(EMOTIONS):
text = str(
round(Decimal(result[0][index] / total_sum * 100),
2)) + "%"
# for drawing progress bar
cv2.rectangle(frame, (100, index * 20 + 10),
(100 + int(result[0][index] * 100),
(index + 1) * 20 + 4), (255, 0, 0), -1)
# for putting emotion labels
cv2.putText(frame, emotion, (10, index * 20 + 20),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (7, 109, 16), 2)
# for putting percentage confidence
cv2.putText(
frame, text,
(105 + int(result[0][index] * 100), index * 20 + 20),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 0, 0), 1)
# overlay emoji on the frame for all the channels
for c in range(0, 3):
# for doing overlay we need to assign weights to both foreground and background
foreground = emoji_face[:, :,
c] * (emoji_face[:, :, 3] / 255.0)
background = frame[350:470, 10:130,
c] * (1.0 - emoji_face[:, :, 3] / 255.0)
frame[350:470, 10:130, c] = foreground + background
break
# Display the resulting frame
cv2.imshow('Video', frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
# When everything is done, release the capture
video_capture.release()
cv2.destroyAllWindows()
def realtime_emotions():
# load keras model
model = define_model()
model = model_weights(model)
print('Model loaded')
# for knowing whether prediction has started or not
once = True
# load haar cascade for face
faceCascade = cv2.CascadeClassifier('haarcascade_frontalface_default.xml')
# list of given emotions
EMOTIONS = ['anger', 'happy', 'neutral', 'sad']
# store the emoji coreesponding to different emotions
emoji_faces = []
for index, emotion in enumerate(EMOTIONS):
emoji_faces.append(cv2.imread('emojis/' + emotion.lower() + '.png', -1))
# set video capture device , webcam in this case
video_capture = cv2.VideoCapture(0)
video_capture.set(3, 640) # WIDTH
video_capture.set(4, 480) # HEIGHT
# save current time
prev_time = time.time()
countt = 0
# start webcam feed
while True:
# Capture frame-by-frame
ret, frame = video_capture.read()
# mirror the frame
frame = cv2.flip(frame, 1, 0)
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
# find face in the frame
faces = faceCascade.detectMultiScale(
gray,
scaleFactor=1.1,
minNeighbors=5,
minSize=(30, 30),
flags=cv2.CASCADE_SCALE_IMAGE
)
# Draw a rectangle around the faces
for (x, y, w, h) in faces:
# draw a rectangle bounding the face
cv2.rectangle(frame, (x-10, y-70),
(x+w+20, y+h+40), (15, 175, 61), 4)
curr_time = time.time()
if curr_time - prev_time >= 1:
roi_gray=gray[y:y+w,x:x+h] #cropping region of interest i.e. face area from image
roi_gray = cv2.resize(roi_gray, (48, 48))
roi_gray = np.reshape(roi_gray, (1, 48, 48, 1))
mean = np.mean(roi_gray)
std = np.std(roi_gray)
roi_gray = (roi_gray-mean)/(std+1e-7)
# do prediction
result = model.predict(roi_gray)
print(EMOTIONS[np.argmax(result[0])])
prev_time = time.time()
total_sum = np.sum(result[0])
# select the emoji face with highest confidence
emoji_face = emoji_faces[np.argmax(result[0])]
for index, emotion in enumerate(EMOTIONS):
text = str(
round(Decimal(result[0][index]/total_sum*100), 2) ) + "%"
# for drawing progress bar
cv2.rectangle(frame, (100, index * 20 + 10), (100 +int(result[0][index] * 100), (index + 1) * 20 + 4),
(255, 0, 0), -1)
# for putting emotion labels
cv2.putText(frame, emotion, (10, index * 20 + 20),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (7, 109, 16), 2)
# for putting percentage confidence
cv2.putText(frame, text, (105 + int(result[0][index] * 100), index * 20 + 20),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 0, 0), 1)
# overlay emoji on the frame for all the channels
for c in range(0, 3):
# for doing overlay we need to assign weights to both foreground and background
foreground = emoji_face[:, :, c] * (emoji_face[:, :, 3] / 255.0)
background = frame[350:470, 10:130, c] * (1.0 - emoji_face[:, :, 3] / 255.0)
frame[350:470, 10:130, c] = foreground + background
break
# Display the resulting frame
cv2.imshow('Video', frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
# When everything is done, release the capture
video_capture.release()
cv2.destroyAllWindows()
