Machine learning on images using a string-distance

TL;DR

This paper introduces a novel image feature-extraction method using the Universal Image Distance to represent images as vectors based on their similarity to prototype images, enabling automatic classification without domain-specific analysis.

Contribution

The paper presents a new, domain-agnostic image feature-extraction technique based on UID, allowing automatic image representation for supervised and unsupervised learning.

Findings

Effective in satellite image classification

No domain knowledge needed for feature extraction

Compatible with various learning algorithms

Abstract

We present a new method for image feature-extraction which is based on representing an image by a finite-dimensional vector of distances that measure how different the image is from a set of image prototypes. We use the recently introduced Universal Image Distance (UID) \cite{RatsabyChesterIEEE2012} to compare the similarity between an image and a prototype image. The advantage in using the UID is the fact that no domain knowledge nor any image analysis need to be done. Each image is represented by a finite dimensional feature vector whose components are the UID values between the image and a finite set of image prototypes from each of the feature categories. The method is automatic since once the user selects the prototype images, the feature vectors are automatically calculated without the need to do any image analysis. The prototype images can be of different size, in particular, different than the image size. Based on a collection of such cases any supervised or unsupervised learning algorithm can be used to train and produce an image classifier or image cluster analysis. In this paper we present the image feature-extraction method and use it on several supervised and unsupervised learning experiments for satellite image data.