Should capture in rgb (not bgr) format.

If `unknown_image.shape[1] > 1600` then `scale_factor` will be always calculated `0` because `1600 / unknown_image.shape[1]` is an `int` over `int` division which in turn will always raise a ValueError in `PIL.Image.resize`. Making `1600` a float (1600.0) fixes the issue.

Replaces test command in .travis.yml so tests get run. Updates
Flake8 configuration. Adds Tox to handle testing of multiple Python
versions. Adds test dependencies to setup.py. Fixes bug in test.

From http://stackoverflow.com/questions/7741878/how-to-apply-numpy-linalg-norm-to-each-row-of-a-matrix

Running time comparisons (prev vs. new, performed using faces = np.random.rand(n, 128), face_to_compare = np.random.rand(1, 128))
n = 1M:  5.3s vs. 1.1s
n = 5M:  26.9s vs. 6.6s
n = 10M: 52.6s vs. 23s

* Refactor Tests

Scipy version fix

Add Pillow as a dependency and bump version