ImageProcessor.py

import base64
import io
import cv2
import numpy as np
from tensorflow.keras.models import load_model
from tensorflow.keras.preprocessing.image import img_to_array
from PIL import Image


def process_image(image_data):

    face_cascade = cv2.CascadeClassifier(
        "./static/xml/haarcascade_frontalface_default.xml")

    model = load_model(
        "./static/models/emotion_detector_1681572728.3088243.h5")
    class_names = {
        0: 'Anger',
        1: 'Disgust',
        2: 'Fear',
        3: 'Happy',
        4: 'Neutral',
        5: 'Sadness',
        6: 'Surprise'
    }

    # Decode the image data
    image_data = base64.b64decode(image_data.split(',')[1])

    # Convert the image data to a numpy array
    nparr = np.frombuffer(image_data, np.uint8)
    img_grey = cv2.imdecode(nparr, cv2.IMREAD_GRAYSCALE)
    normal_img = cv2.imdecode(nparr, cv2.IMREAD_COLOR)

    # Convert normal image to RGB
    normal_img = cv2.cvtColor(normal_img, cv2.COLOR_BGR2RGB)

    # Detect the face in the image
    faces = face_cascade.detectMultiScale(img_grey, 1.3, 5)
    if len(faces) == 0:
        return {'prediction': 'no face detected'}

    # Get the first face detected
    (x, y, w, h) = faces[0]

    # Draw a rectangle around the face
    img_with_rect = cv2.rectangle(
        normal_img.copy(), (x, y), (x+w, y+h), (0, 255, 0), 2)

    # Crop the image to the face
    img_grey = img_grey[y:y+h, x:x+w]

    # Resize the image and convert it to an array
    img_grey = cv2.resize(img_grey, (48, 48))
    img_array = img_to_array(img_grey)
    img_array = img_array.reshape(
        (1, img_array.shape[0], img_array.shape[1], 1))

    prediction = model.predict(img_array)
    predicted_class = np.argmax(prediction)
    predicted_emotion = class_names[predicted_class]

    # Write the predicted emotion on the image
    img_with_text = cv2.putText(img_with_rect, predicted_emotion, (x, y-10),
                                cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 1)

    # Convert the image to a JPEG string
    img_buffer = io.BytesIO()
    Image.fromarray(img_with_text).save(img_buffer, format="JPEG")
    img_str = base64.b64encode(img_buffer.getvalue()).decode()

    response = {'prediction': predicted_emotion,
                'imageData': img_str,
                }

    return response