The independent components of binocular images

The response properties of cells in the visual cortex appear to be optimally tuned to provide a sparse, economic representation of natural images. Independent-component analysis was performed to determine how a similar representation of binocular information might be achieved. Binocular photographs of natural scenes were captured with a binocular video camera. For each photograph, the camera was fixated on a point of central interest in the scene. Independent-component analysis was performed on a large number of samples extracted from the binocular image pairs. The resulting independent-component filters showed clearly defined receptive fields, and were tuned to both orientation and spatial frequency. A substantial proportion of binocular filters, showing similar response properties for the left and right images, were obtained. Close analysis of these filters revealed interocular differences in both the position and structure of their receptive fields. Filters were obtained that differed in phase, orientation, and spatial-frequency-tuning between the left and right images. These results show close similarities to physiological results (DeAngelis et al, 1995 Perception 24 3 - 31; Bridge and Cumming, 2001 Journal of Neuroscience 21 7293 - 7302), and suggest that the visual system might use a number of strategies to encode binocular disparity.