Texture discrimination by cells in the cat lateral geniculate nucleus

Abstract

The spontaneous segregation of texture areas is an impressive perceptual phenomenon, the neural basis of which is not yet understood. In the texton concept (Julesz and Bergen 1983; Julesz 1984, 1986) it is assumed that the visual system analyzes a stimulus for certain features ('textons') the spatial distribution of which is pre-attentively registered and may provide the percept of dissected texture areas. Supposed textons are blobs of a given size, oriented lines, line intersections and line terminators, suggesting that texture analysis is exclusively mediated by form-specific filters at higher, e.g. cortical, processing levels. This paper investigates the contribution of cells in the cat lateral geniculate nucleus (LGN) to segregation of typical texton differences. The results indicate that LGN cells, though not resembling the supposed texton filters, often distinguished textured arrangements of such features on the basis of a variety of other visual cues, such as global or local variations in mean luminance or differences in spatial frequency composition. Thus, cells responded to texture borders between areas differing in the size or the density of texture elements and often revealed differential firing rates to textures differing by the crossing or the terminator feature. For textures with differences in line orientation, however, only small variations of the firing rate were seen. In summary, the observations suggest a means of texture representation in the cat LGN which is different from recent concepts of texture segregation in man. For a given pair of textures, cells with receptive fields larger than, or similar to the texture raster respond to global and local luminance variations between areas and, in particular, to differences in their spatial frequency composition. These cells, hence, may signal the global texture difference without encoding spatial details of the pattern from which texton features could be identified. Cells with receptive fields small in comparison to texture elements transfer all the information necessary for analyzing these elements in detail, but themselves are relatively insensitive to global texture differences.