 |
Center for Nonlinear Studies |
Characterizing the coding strategies used by a sensory system is a fundamental goal in neuroscience research. Attaining this goal has particularly important implications for applied sciences, which can translate neural coding strategies into models for solving vexing problems such as automated object recognition. In the realm of visual processing, much of the existing experimental work has focused on characterizing the coding principles of a visual pathway originating at the retina known either as the X-cell or parvocellular pathway. The responses of X-cells are dominantly linear and well-modeled using linear spatiotemporal filters similar to those at the front-end of many computer vision systems. There exists a second major visual pathway originating at the retina known either as the Y-cell or magnocellular pathway. The responses of Y-cells are known to be nonlinear; however, the principles by which they encode the visual scene are poorly understood. Identifying these principles is a necessary first step toward translating the Y-cell pathway into a model with practical applications involving computer vision. As such, the goal of the present study is to functionally identify the nonlinear coding scheme underlying the responses of Y-cells. This work advances a recent publication by the presenter demonstrating that Y-cells account for nonlinear response properties previously measured in and thought to arise in cortex (Rosenberg et al., 2010). Electrophysiological recordings of individual Y-cells were made in the cat lateral geniculate nucleus, the subcortical target through which visual information is passed from the retina to cortex. Using techniques from Fourier analysis, the nonlinear coding scheme of the Y-cell pathway is identified in the present study as demodulation, a nonlinear engineering principle with numerous applications including the decoding of AM radio signals. It is shown that the Y-cell pathway is in many regards anatomically and functionally similar to the demodulating circuit in an AM radio receiver. Although the importance of visual demodulation is not well understood at this time, existing theoretical work suggests that demodulation is an efficient way to encode texture patterns important for pattern recognition and boundary detection. Examples demonstrating this will be presented. In addition, the present results have important implications relating visual acuity and the manner in which temporally varying stimuli are encoded in the visual system. Having identified the nonlinear coding scheme as demodulation allows for straightforward modeling of the Y-cell visual pathway. Such a model will not only provide a way to systematically study the role of this nonlinear coding scheme in visual processing but also enable solutions to real-world image processing problems.
Rosenberg A., Husson, T.R., & Issa, N.P. Subcortical representation of non-Fourier image features. Journal of Neuroscience, 30(6): 1985-1993, 2010.