A New Scheme for Color Night Vision by Quaternion Neural Network

A neural network for extracting color information from gloomy images is presented in this paper. We adopt a feedforward quaternion neural network, where neuronal parameters are represented as 4-dimensional vectors. The network is trained by imposing a gloomy image as input and its finer one as target by quaternion version of back propagation algorithm, and then relations between them are established. We show the performance and the generalization ability of quaternion neural network through numerical results. Our proposed method will be useful in realizing a so-called color night vision system for the benefit of its simplicity and cost-effectiveness.