Perceptually Inspired Normalized Conditional Compression Distance

TL;DR

This paper introduces a perceptually-inspired extension of the Normalized Information Distance for image similarity, leveraging the Divisive Normalization Transform to better model human visual perception in various image processing tasks.

Contribution

It proposes a novel Normalized Conditional Compression Distance that incorporates human visual system modeling for improved image similarity measurement.

Findings

Effective in texture classification

Improves face recognition accuracy

Applicable across diverse image processing tasks

Abstract

Image similarity measurement is a common issue in a broad range of applications in image processing, recognition, classification and retrieval. Conventional image similarity measures are often limited to specific applications and cannot be applied in general scenarios. The theory of Kolmogorov complexity provides a universal framework for a generic similarity metric based on information distance between objects. Normalized Information Distance (NID) has been shown to be a valid and universal distance metric applicable in measurement of similarity of any two objects, and has been successfully applied to a wide range of applications in the past. The difficulty of NID lies in the non-computable nature of the Kolmogorov complexity, and thus approximation has to be applied in practice. Here we propose a perceptually-inspired Normalized Conditional Compression Distance (NCCD) measure by using the Divisive Normalization Transform (DNT) as a means to model the non-linear behavior of the Human Visual System (HVS) in reducing statistical dependencies of visual signals for efficient representation, and show that this perceptual extension of NID can be used in a wide range of image processing applications, including texture classification and face recognition.