Purchase this article with an account.
Qin Wang, Isamu Motoyoshi, Shin’ya Nishida; Characterization of high-level images features for surface gloss perception. Journal of Vision 2013;13(9):202. doi: 10.1167/13.9.202.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Effective cues for surface gloss perception reside in low-level image statistics, including skewness of luminance or subband histogram (Motoyoshi, Nishida, Sharan & Adelson, 2007). It is also suggested that gloss perception is modulated by high-level spatial information such as consistency between patterns of specular highlight and diffuse shading (Anderson & Kim, 2009). To gain further insight into the nature and the processing mechanism of high-level gloss features, we examined the effects of highlight consistency, peripheral viewing and texture synthesis on gloss perception. The stimulus we used was a surface modulated in depth by a low-pass-filtered random field and rendered by the Ward illumination model. We made a highlight-consistent image by combining specular and diffuse patterns made from the same depth profile, and a highlight inconsistent image by combining specular and diffuse patterns made from different uncorrelated depth profiles. Using the algorithm by Portilla & Simoncelli (2000), we also synthesized two types of texture image each sharing low-level image statistical measurements with the highlight-consistent and highlight-inconsistent surface images, respectively. We used a magnitude rating, and a pairwise magnitude comparison, to evaluate gloss perception for these images. The results indicate that highlight-consistent surfaces looked much more glossy than highlight-inconsistent surfaces when viewed in the fovea, but this difference was reduced as the retinal eccentricity was increased. Synthesized textures did not look glossy at any eccentricity, and it was hard to discriminate highlight-consistent textures from inconsistent ones. Our findings suggest that critical image features that produce the apparent gloss difference between highlight-consistent and inconsistent images are available to human observers mainly through foveal vision, and that these features cannot be captured by the image statistics preserved by the Portilla & Simoncelli algorithm.
Meeting abstract presented at VSS 2013
This PDF is available to Subscribers Only