Hands-On Image Processing with Python

k-nearest neighbors (KNN) classifier

In this section, we will build a classifier that takes an image of a handwritten digit and outputs a label (0-9) using a particularly simple strategy for this problem known as the nearest neighbor classifier. The idea to predict an unseen test digit image is pretty simple. First, we need to find the k instances from the training dataset that are nearest to this test image. Next, we need to simply use majority-voting to compute the label of the test image, that is, the label that most of the data points from the k nearest training data points have will be be assigned to the test image (breaking ties arbitrarily).