This repo contains the code for Image Blur Classifcation. Two methods have been used to solve this problem:-

Pre-requisites

The code is developed for Python 2.7 using Jupyter notebook

All dependencies are mentioned in requirements.txt

Laplacian Method

This method is based on this tutorial. The test accuracy achieved using this method for :-

1.NaturalBlurSet -> ~77% 2.DigitalBlurSet -> ~96%

Deep Learning based Method

This method uses Convolution Neural Network for the classification. The test accuracy achieved using this method for :-

1.NaturalBlurSet -> ~61% 2.DigitalBlurSet -> ~62%

Training

The training was done on Google Colaboratory. The training data was saved in pickle files (generated using files pre-processing directory) and uploaded on the Cloud. The final trained model weights were downloaded and tested againsta the evaluation set.

Dataset

It is trained on CERTH Image Blur Dataset