Does the human visual system implement an ideal observer theory of slant from texture?

Texture information about surface shape can be decomposed into three constituents: compression, density and scaling. Blake, Bülthoff and Sheinberg's (1993, Vision Research, 33, 1723-1737) Ideal Observer theory of slant from texture predicts that the relative contribution of compression and density to the percept of planar surface slant should vary with field of view (FOV). The contribution of compression and density should both increase as FOV increases but statistical analysis shows that the reliability, and hence the expected contribution, of density increases at a greater rate than compression with increasing FOV. Specific predictions are that at FOV < 20 deg compression should be more effective than density, at FOV approximately 20 deg, compression and density should be equally effective, and at FOV > 20 deg, compression should be less effective than density. These predictions were tested by pitting these components of texture against one another. The method used was similar to that described in Frisby and Buckley [1992 In Orban, G. & Nagel H.-H. (Eds) Artificial and biological visual systems. Berlin: Springer]: observers judged binocularly the slant of a large table on to which was projected a texture created using computer graphics to contain various texture component cues to slant. The size of the projected texture patch determined the FOV which was by this means set to either 10, 20 or 30 deg. It was found that compression was the dominant cue irrespective of FOV, even if it was perturbed by noise. Possible reasons for this divergence of human observers from predictions of the Ideal Observer theory of slant from texture are discussed.