For SVHN Dataset- http://ufldl.stanford.edu/housenumbers/ (Format 2 - train_32x32.mat, test_32x32.mat,extra_32x32.mat)

In python

1. Preprocess it . executing it will generate a set of processed files.

2. Visualization - treat images as numbers. They are just matrices computer percentage of image that has dark vs light pixels, average color of each image, the number of digits that occur in image etc. Executing it will generate all of the visuals for your data, provided processed data files already exist

3. after doing visualization, for that find correlations.

use libraries only

- tensorflow

- matplotlib

- numpy

- pandas

- requests

- beautifulsoup4

- NLTK