Perception of super-fine structures based on image intensity statistics

We feel that we can recognize the fineness of surface textures with very fine structures, such as human hair, either directly or through photographic images, even when the spatial scale of their individual elements is finer than the spatial resolution limit of the visual system or the physical resolution of the digitized image. Fineness perception of texture might rely not only on the spatial-frequency information of the texture, but also on other diagnostic image features. To investigate this possibility, we first explored to what extent human observers can distinguish differences in super-fine structure. We made a multi-resolution sequence of one-dimensional hair-like random textures through successively applying low-pass filtering and down-sampling. Results of the pairwise comparison between these textures showed that observers could correctly evaluate the fineness of the textures even when the thinnest element was much thinner than the resolution limit of the visual system or that of the digitized image. What happened in the image? According to the central limit theorem, as the fineness of texture increases and the number of elements per pixel increases, the intensity contrast of the texture decreases and the intensity histogram approaches a Gaussian shape. Subsequent experiments revealed that these image features indeed play critical roles in the fineness perception. Specifically, for textures with a unimodal (e.g., Gaussian) distribution, observers perceived the contrast-reduced texture to be finer than the original one. In comparison, for textures with a bimodal distribution, contrast reduction had little effect on fineness perception. These findings suggest that the visual system utilizes the intensity contrast of the texture for estimation of the magnitude of fineness (lower contrast makes the texture look finer) under the condition that the shape of the intensity distribution is consistent with the characteristics of super-fine textures. Meeting abstract presented at VSS 2016