CNNs_street-address-recognition

MLND Deep Learning Capstone Project to detect Street View House Number using tensorflow, deep learning platform

In this project, I would decode and recognize the sequences of digits from the natural images of Street View House Numbers (SVHN) through training Convolutional neural networks (CNN), which is the special case of the neural network with convolutional layers and subsampling layers. This model could enable us to find the housing number of a specific location in a street as a format of continuous multiple digit characters with 94.7% prediction accuracy from test data, which is shy of human recognition 97% but could be a good starting point for us to improve to excel the capability of human vision recognition.

• In order to exute the include program, First of all ipython notebook or jupyter notebook should be installed
• This program is developed and test on GTX1080.

Source Code

• capstone_preparation_project.ipynb
• Capstone_digitStruct.ipynb
• capston_digitStructMatToCsv.ipynb
• capstone_main_preprocess_project.ipynb
• capstone_main_CNN_project

Requirements

• Python 2.7
• Jupyter Notebook
• TensorFlow
• NumPy
• matplotlib
• scikit-learn
• SciPy library
• Six
• h5py

This project files are tested and optimized 32GB intel Core7 system with GTX1080 GPU.

You will also need to have software installed to run and execute an Jupyter Notebook`

Run

Open Ipython Notebook in the root folder by type “jupyter notebook” on the shell command line(linux) or command line(windows) Open the 5 Files and run each cell one by one in the files.

Data

This project uses the The Street View House Numbers (SVHN) Dataset These dataset are downloaded and extracted in capstone_preparation_project.ipynb and capstone_main_preprocess_project.ipynb