image_cv.py

import cv2
import numpy as np
from keras.models import load_model
from PIL import Image, ImageOps


def cv2_to_pil(img): #Since you want to be able to use Pillow (PIL)
    return Image.fromarray(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))

def pil_to_cv2(img):
    return cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR)

# Load the model
model = load_model('keras_model.h5')
label = {0: '50k', 1: '100k'}
size = (224, 224)
data = np.ndarray(shape=(1, 224, 224, 3), dtype=np.float32)

img = cv2.imread('data/100k2.jpg', cv2.IMREAD_COLOR)
resized_img = cv2.resize(img, size)
# cv2.imshow('img',img)

image = cv2_to_pil(img)
image = ImageOps.fit(image, size, Image.ANTIALIAS)
# turn the image into a numpy array
image_array = np.asarray(image)
# Normalize the image
normalized_image_array = (image_array.astype(np.float32) / 127.0) - 1
# Load the image into the array
data[0] = normalized_image_array

# run the inference
prediction = model.predict(data)
p = np.argmax(prediction)
print ('prediction',prediction)

cv2.putText(resized_img, 'result: {}'.format(label[p]), (20, 50), cv2.FONT_HERSHEY_DUPLEX, 1, (0, 255, 255), 1)
cv2.imshow('prediction', resized_img)
cv2.waitKey(0)