A Simple CW-SSIM Kernel-based Nearest Neighbor Method for Handwritten Digit Classification

TL;DR

This paper introduces a kernel-based nearest neighbor method using CW-SSIM for handwritten digit classification, achieving near state-of-the-art accuracy on MNIST with minimal neighbors.

Contribution

It presents a simple kernel-based approach that effectively measures image similarity for digit classification, demonstrating competitive accuracy with few neighbors.

Findings

Test error rate of 1.5%-2.0% on MNIST

Effectiveness of different neighbor counts and weight schemes

Close performance to advanced models with simple method

Abstract

We propose a simple kernel based nearest neighbor approach for handwritten digit classification. The "distance" here is actually a kernel defining the similarity between two images. We carefully study the effects of different number of neighbors and weight schemes and report the results. With only a few nearest neighbors (or most similar images) to vote, the test set error rate on MNIST database could reach about 1.5%-2.0%, which is very close to many advanced models.