Eigen-Distortions of Hierarchical Representations

TL;DR

This paper introduces a method using Fisher information to compare hierarchical image representations based on their ability to predict human perceptual sensitivity to local image distortions.

Contribution

It develops a novel approach to evaluate and compare neural network layers and models in terms of their alignment with human perceptual sensitivity using eigen-distortions.

Findings

Early VGG16 layers better match human perception than later layers.

Simple models with local gain control outperform CNNs in predicting human sensitivity.

Eigen-distortions effectively reveal differences in perceptual relevance of image features.

Abstract

We develop a method for comparing hierarchical image representations in terms of their ability to explain perceptual sensitivity in humans. Specifically, we utilize Fisher information to establish a model-derived prediction of sensitivity to local perturbations of an image. For a given image, we compute the eigenvectors of the Fisher information matrix with largest and smallest eigenvalues, corresponding to the model-predicted most- and least-noticeable image distortions, respectively. For human subjects, we then measure the amount of each distortion that can be reliably detected when added to the image. We use this method to test the ability of a variety of representations to mimic human perceptual sensitivity. We find that the early layers of VGG16, a deep neural network optimized for object recognition, provide a better match to human perception than later layers, and a better match than a 4-stage convolutional neural network (CNN) trained on a database of human ratings of distorted image quality. On the other hand, we find that simple models of early visual processing, incorporating one or more stages of local gain control, trained on the same database of distortion ratings, provide substantially better predictions of human sensitivity than either the CNN, or any combination of layers of VGG16.