Recovery of Lost Color Information Using Neural Network

Abstract

We propose a method of recovering lost information in an image using a neural network. As human brains recall a missing piece of information from memory of multiple events that occur simultaneously, the lost information is recovered from typical combinations of colors memorized in the network. We show by experiments that one of color channels is recovered with good fidelity from remaining information using a neural network as simple as the self-organizing map. Even if our success is limited to images with only a few colors, the same idea would apply to recovery of more useful information as the network evolves to be more complicated.