Building Machine Learning Systems with Python - Richert, Coelho

Chapter 10
The result is that each image is now represented by a single array of features of
the same size (the number of clusters; in our case 256). Therefore, we can use our
standard classification methods. Using logistic regression again, we now get 62
percent, a 7 percent improvement. We can combine all of the features together and
we obtain 67 percent, more than 12 percent over what was obtained with texturebased
methods:
Summary
We learned the classical feature-based approach to handling images in a machine
learning context by reducing a million pixels to a few numeric dimensions. All
the technologies that we learned in the other chapters suddenly become directly
applicable to image problems. This includes classification, which is often referred
to as pattern recognition when the inputs are images, clustering, or dimensionality
reduction (even topic modeling can be performed on images, often with very
interesting results).
We also learned how to use local features in a bag-of-words model for classification.
This is a very modern approach to computer vision and achieves good results while
being robust to many irrelevant aspects of the image, such as illumination and
also uneven illumination in the same image. We also used clustering as a useful
intermediate step in classification rather than as an end in itself.
[ 219 ]