Handwritten_Digits_Recognition

A web service of recognizing digits on image.

• Build Image Preprocess Pipeline.

  • Convert to GrayScale

    

  • Apply Gaussian Blur

    

  • Apply threshold and invert colors

    

  • Use Canny Edge Detection to find edges

    

  • Dilate Edges found

    

  • Find Contours and bouding box for each digit

  • Extract each digit with boarder added to improve performance

    

• Import MNIST digits data.

• Build and Train a Deep Neural Network Model to predict digits based on image pixels.

  • Convolutional Neural Network Pipeline:

  • Model performance

    # calculate accuracy on the prediction
    acc = np.average(Y_pred == Y_true)
    print('Accuracy is', acc)
    Accuracy is 0.9912857142857143
    Attempted to log scalar metric accuracy:
    0.9912857142857143

• Store the DNN model into an ONNX:Open Neural Network Exchange instance.

• Load and Register the ONNX model into Azure.

• Create a Docker Image of the registered model, score.py script, and YAML envrionment dependencies file.

• Deploys the scoring image on Azure Containter Instance (ACI) as a web service.

• Client convert image into bytes array and serialize it into JSON str.

• Client send a HTTP request with JSON str to digits recognition web service.

• The web service decode JSON str into bytes array which is sent to CNN model to predict digits on the image.

• The prediction of digits are returned to the client.

  import os
  import json
  import base64
  import requests
  
  image_path = os.path.join(os.getcwd(), 'test1.jpg')
  with open(image_path, 'rb') as file:
    img = file.read()
  image_64_encode = base64.encodebytes(img).decode('utf-8')
  bytes_to_json = json.dumps(image_64_encode)
  
  scoring_url = 'http://3ab34ad2-281d-4017-b47f-7a099895a46b.centralus.azurecontainer.io/score'
  headers = { 'Content-Type': 'application/json' }
  response = requests.post(scoring_url, bytes_to_json, headers=headers)
  print(json.loads(response.text))
  [0, 7, 1, 3, 8, 9]