Long-term adaptation mechanisms for fine-tuning of man-made sensory processing systems

摘要

Neurobiological systems possess a tremendous ability to adapt to the surrounding environment at multiple time scales and at multiple stages of processing. Though the purpose of these biological adaptation mechanisms is not clear, some theories suggest that these methods allow for the fine-tuning of the visual system through long-term averaging of measured visual parameters. We have developed the constant statistics model to apply these biologically plausible adaptation constraints to the design of man-made sensory systems. This paper discusses examples of long-term adaptation in the nervous system and shows how similar constraints can be exploited in man-made sensory processing systems. We review several examples of such biologically inspired adaptation mechanisms for such engineering problems as offset/gain correction of IR imagers, adaptive signal processing, motion estimation, and preliminary results in optimal scale detection in image processing.