IEEE Access (Jan 2020)

A Convolutional Neural Network Reaches Optimal Sensitivity for Detecting Some, but Not All, Patterns

  • Fabian Hubert Reith,
  • Brian A. Wandell

DOI
https://doi.org/10.1109/ACCESS.2020.3040235
Journal volume & issue
Vol. 8
pp. 213522 – 213530

Abstract

Read online

We investigate the spatial contrast-sensitivity of modern convolutional neural networks (CNNs) and a linear support vector machine (SVM). To measure performance, we compare the CNN contrast sensitivity across a range of patterns with the contrast sensitivity of a Bayesian ideal observer (IO) with the signal-known-exactly and noise-known-statistically. A ResNet-18 reaches optimal performance for harmonic patterns, as well as several classes of real world signals including faces. For these stimuli the CNN substantially outperforms the SVM. We further analyze the case in which the signal might appear in one of multiple locations and found that CNN spatial sensitivity continues to match the IO. However, the CNN sensitivity is far below optimal at detecting certain complex texture patterns. These measurements show that CNNs spatial contrast-sensitivity differs markedly between spatial patterns. The variation in spatial contrast-sensitivity may be a significant factor, influencing the performance level of an imaging system designed to detect low contrast spatial patterns.

Keywords