# MNIST real-time webcam辨識程式
# prg_CNN_webcam
# Author: Wu, Chih-Hung
# 2022/12/7
# 原始出處: 高怡宣老師
# 修改自上述程式
# 雲端空間
# https://drive.google.com/drive/folders/1o7u9yEnRE4FmPDbUwwGV2bu0lX_lDHeI
# Youtube
# https://www.youtube.com/watch?v=CsJH1wvgJ28

import matplotlib.pyplot as plt
import matplotlib.image as mpimg
import numpy as np
import cv2
import time

from keras.models import load_model
model = load_model('my_model.h5')
#model.summary()
model.load_weights('my_model_weights.h5')

#如果 load_model 有錯誤訊息, 請檢查 keras 和 tensorflow 的版本.
#建議使用 keras 2.2.4, tensorflow 1.8.0 or 1.12.0

cap = cv2.VideoCapture(0)
width = cap.get(cv2.CAP_PROP_FRAME_WIDTH)
height = cap.get(cv2.CAP_PROP_FRAME_HEIGHT)
print("Image Size: %d x %d" % (width, height))

while(True):
    tStart = time.time() #計時開始
    ret, img_1 = cap.read()
    img_2 = img_1[0:480,80:560]
    cv2.imshow('Camera',img_2)
    
    img_2_gray = cv2.cvtColor(img_2, cv2.COLOR_BGR2GRAY)
    img_3 = 255 - img_2_gray
    img_3 = img_3.astype('float32')
    img_3_min = np.amin(img_3)
    img_4 = img_3 - np.amin(img_3)
    img_5 = 255 * img_4 / (np.amax(img_4))
    kernel = np.ones((5,5),np.uint8)
    img_6 = cv2.dilate(img_5,kernel,iterations = 3)
    img_7 = cv2.resize(img_6,(28,28),1)
    img_8 = img_6.astype('uint8')
    cv2.imshow('input',img_8)

    #print(img_1.dtype, img_1.shape)
    #print(img_2.dtype, img_2.shape)

    x_test_image = np.reshape(img_7, (1,28,28))
    #print(np.amax(x_test_image))
    #print(np.amin(x_test_image))

    # convert 2-D 28x28 image to 4-D nx28x28x1  array
    x_Test4D=x_test_image.reshape(x_test_image.shape[0],28,28,1).astype('float32')
    # normalize the image numbers to 0~1
    x_Test4D_normalize = (x_Test4D / np.amax(x_test_image))
    #print(x_Test4D_normalize.shape)
    
    # 原始程式需修正的地方
    # 1. 修改model.predict_classes -> model.predict
    # 2. model.predict()內的預測資料變數修改為 -> x_Test4D_normalize
    
    #舊版呼叫方式
    #prediction=model.predict_classes(x_Test4D_normalize)
    
    prediction = (model.predict(x_Test4D_normalize))
    prediction = np.argmax(prediction,axis=1)
    prediction
    
  


    filename = '%s%d%s' % ('./numbers/number_', prediction, '.jpg')
    img_9 = cv2.imread(filename)
    img_10 = cv2.resize(img_9,(480,480),1)
    cv2.imshow('inference',img_10)
    tEnd = time.time() #計時結束

    if cv2.waitKey(1) & 0xFF == ord('q'):
        cv2.imwrite('output.jpg', img_2)
        break

cap.release()
cv2.destroyAllWindows()

print('It takes %f sec for each frame' % (tEnd - tStart))