|
from utils import *
|
|
from flask import Blueprint,render_template,request,jsonify
|
|
import tensorflow as tf
|
|
import base64
|
|
import io
|
|
from PIL import Image
|
|
views = Blueprint("views",__name__)
|
|
|
|
ocr_model = None
|
|
ocr_sub_model = None
|
|
|
|
|
|
MODEL_PATH = 'models/ocr_big_1.h5'
|
|
SUB_MODEL_PATH= 'models/ocr_new_1.h5'
|
|
|
|
@views.route('/',methods=['GET','POST'])
|
|
def index():
|
|
global ocr_model
|
|
global ocr_sub_model
|
|
|
|
if request.method == 'POST':
|
|
data = request.json['image']
|
|
head, data = data.split(',', 1)
|
|
image_data = base64.b64decode(data)
|
|
|
|
image = Image.open(io.BytesIO(image_data)).convert('RGB')
|
|
|
|
processed_image = np.array(image)
|
|
|
|
if not ocr_model:
|
|
ocr_model = tf.keras.models.load_model(MODEL_PATH)
|
|
if not ocr_sub_model:
|
|
ocr_sub_model = tf.keras.models.load_model(SUB_MODEL_PATH)
|
|
|
|
|
|
detected_contours,output_image, coords = get_results(processed_image)
|
|
output_string = ""
|
|
for i in range(len(detected_contours)):
|
|
predicted_char = predict(ocr_model, ocr_sub_model,detected_contours[i],mappings)
|
|
|
|
if predicted_char == '0':
|
|
output_string += 'O'
|
|
cv2.putText(output_image, 'O', (coords[i][0]+5,coords[i][1]-5), font, font_scale, color, thickness)
|
|
else:
|
|
cv2.putText(output_image, predicted_char, (coords[i][0]+5,coords[i][1]-5), font, font_scale, color, thickness)
|
|
output_string += predicted_char
|
|
|
|
|
|
output_image = Image.fromarray(output_image, 'RGB')
|
|
data = io.BytesIO()
|
|
output_image.save(data, "JPEG")
|
|
output_image = base64.b64encode(data.getvalue()).decode('utf-8')
|
|
return jsonify({'output_image': output_image, 'output_string': output_string})
|
|
return render_template('index.html')
|
|
|
|
|
