This special issue represents the third devoted to natural scene statistics in this Journal. It was spurred by the Gordon Conference on Sensory Coding and the Natural Environment held 27 August to 1 September, 2006 in Big Sky Montana. This meeting brought together neuroscientists, psychologists, mathematicians, and engineers who seek to characterize the structure of natural scenes (including images, sound and other senses) or who study how this information is coded and represented in the brain. The larger theme of the meeting was to understand how biological systems encode and process complex, natural stimuli under natural conditions. Participants of this meeting and the scientific community at large were invited to submit manuscripts for this special issue of Network. The papers in the issue were fully peer-reviewed and represent a sample of the broad range of topics in this field.
Why study natural scenes? Although simple signals are easy to understand and model, systems capable of processing simple signals often fail when presented with natural scenes. The inherent complexity of any system capable of processing natural scenes is reflected in neural sensory systems, which through evolution have become exquisitely attuned to the statistical structure of natural stimuli. Given the daunting complexity of both our natural sensory environment and the neural circuitry of our sensory systems, it was remarkable when (many decades ago now) physiologists first discovered that visual neurons responded to simple features like spots and bars. Probing and characterizing neural responses with simple stimuli was one of the catalyzing methodologies in the development of the field of neuroscience. However, the simple stimulus approach lacked a theoretical framework that could bridge the gap between simple stimuli and natural scenes and provide deeper insight into the nature of neural processing.
In 1987, David Field published a seminal paper showing that primary visual cortex (area V1) is optimally adapted to process natural images. It had long been hypothesized that sensory systems evolved mechanisms matched to the statistical structure of natural signals, but before that time it was generally believed that natural signals were too complicated to be useful in most experimental or theoretical studies of sensory processing. With this result and advances in computing technology (which enabled the development of increasingly sophisticated hypotheses and experiments) the study and use of natural scenes has advanced rapidly.
Natural scenes have become a powerful tool for making predictions from theoretical models, and for experimental investigation of neural representation and processing. Papers on natural scenes are published routinely in leading journals in several theoretical and experimental fields. Presentations at the last conference on Sensory Coding and the Natural Environment covered not only the traditional subjects such as natural image statistics and visual representation, but also the use of natural images to investigate higher-level visual areas, theoretical and experimental studies on natural sounds, natural scene perception in humans, adaptation and learning, and active sensing and learning. The field continues to expand and promises to yield deeper insight in the principles of biological perception and function. The next Gordon Research Conference on “Sensory Coding and the Natural Environment” will be held during 27 July to 1 August, 2008 at Il Ciocco, Barga, Italy.