Probe: Noise-and-rotation resistance of Hopfield Neural Network in imaged traffic sign recall

Abstract

This paper examines the noise and rotation resistance capacity of Hopfield Neural Network (HNN) given four corrupted traffic sign images. In the study, Signal-to-Noise Ratio (SNR), recall rate and pattern complexity are defined and employed to evaluate the recall performance. The experimental results indicate that the HNN possesses significant recall capacity against the strong noise corruption, and certain restoring competence to the rotation. It is also found that combining noise with rotation does not further challenge the HNN corruption resistance capability as the noise or rotation alone does.