Optical EngineeringColor image enhancement using a multiple-scale opponent neural network
|Format||Member Price||Non-Member Price|
One goal of image processing is to recreate the human visual perception of the original scene. We offer a method of color image enhancement for realistic image display. The algorithm is based on a special implementation of the feed-forward ON-OFF shunting neural network. The properties of the neural network for image processing are first investigated. The computational simulations make clear the strengths and weaknesses of the neural network. Combined with a brightness-based tone reproduction operator, a multiscale extension to the neural network is developed to correct the original version's weaknesses and provides a coherent framework for color image enhancement. The algorithm achieves both dynamic range compression and vivid color rendition and qualitatively approaches the aspects of early visual perception. Extensive tests demonstrate that this method could produce satisfying images without any unexpected results.