Human and DNN classification performance on images with quality distortions: A comparative study

Samuel Dodge, Lina Karam

Research output: Contribution to journalArticlepeer-review

28 Scopus citations

Abstract

Image quality is an important practical challenge that is often overlooked in the design of machine vision systems. Commonly, machine vision systems are trained and tested on high-quality image datasets, yet in practical applications the input images cannot be assumed to be of high quality. Modern deep neural networks (DNNs) have been shown to perform poorly on images affected by blur or noise distortions. In this work, we investigate whether human subjects also perform poorly on distorted stimuli and provide a direct comparison with the performance of DNNs. Specifically, we study the effect of Gaussian blur and additive Gaussian noise on human andDNNclassification performance.We perform two experiments: one crowd-sourced experiment with unlimited stimulus display time, and one lab experiment with 100ms display time. In both cases, we found that humans outperform neural networks on distorted stimuli, even when the networks are retrained with distorted data.

Original languageEnglish (US)
Article number7
JournalACM Transactions on Applied Perception
Volume16
Issue number2
DOIs
StatePublished - Mar 2019

Keywords

  • Deep learning
  • Human study
  • Robust visual recognition

ASJC Scopus subject areas

  • Theoretical Computer Science
  • General Computer Science
  • Experimental and Cognitive Psychology

Fingerprint

Dive into the research topics of 'Human and DNN classification performance on images with quality distortions: A comparative study'. Together they form a unique fingerprint.

Cite this