Segmentation of image cues for perceived gloss of grapes in painted still lifes

Understanding which variables are involved in gloss judgment is a widely-investigated topic in the field of material perception. Although several mechanisms have been suggested, there is still considerable disagreement about the identity and the complex interaction of the elements playing a role. Here, we proposed a new approach to analyse lowlevel image features, namely highlights’ contrast and sharpness, contributing to gloss perception. Psychophysical experiments were conducted using as stimuli 81 digital images of 17th century paintings, depicting grapes. A group of 10 observers was asked to rate the perceived glossiness of the grapes. The task of another group of 7 observers was to segment, from the same set of images, the cues they considered to be responsible for gloss perception. All participants regarded the highlights as the most salient cue. Thereafter, highlights’ contrast and sharpness were calculated for each segmented grape, via computation of Michelson’s contrast and steepness, from their intensity profiles. A strong linear correlation was found between perceived gloss and highlights’ contrast (ρ=0.65, p<0.01), whereas sharpness showed a slightly lower correlation (ρ=0.39, p<0.01). One advantage of our method was that observers were able to freely assess gloss cues, which could then be directly measured from the images. Moreover, the choice of paintings as stimuli demonstrated that the study of perceived gloss is not restricted to controlled computer-rendered images. Our findings validated the proposed role of highlights’ contrast and sharpness, as part of the aspects to take into account when computing gloss perception. Facial identification and feature integration under memory load Kaisu Ölander, Toni Saarela, Ilkka Muukkonen & Viljami Salmela University of Helsinki We studied working memory precision for facial identities and the integration of facial features with and without memory load. Our stimuli consisted of face images of 60 identities from the Radboud and FACES databases. We formed 20 continuous, circular identity spaces by morphing subsets of the faces from one identity to another. On each trial, participants adjusted a probe face to match a target identity either while viewing the original target stimulus (perception) or after a 1.5 s retention period while keeping in mind 1-4 identities (memory). Under memory load, the target face was indicated with a spatial cue after the retention period. We measured adjustment-error distributions for eye region only, mouth region only, and the whole face. We estimated the precision of perception/memory from von Mises distributions fit to these data. As expected, precision was higher in the perception than in the memory condition and also for the whole face compared to mouth or eyes alone. From the single-feature conditions (eyes and mouth), we estimated the noise in the coding of individual features and the additional noise due to the memory load. We then predicted the whole-face precision assuming optimal integration of independent features. The data closely followed the model predictions. We conclude that memory precision declines with increasing memory load similarly for individual features and for the whole face. The near-optimal precision for the whole face suggests that the decline is mainly due to memory noise, not due to disruption of feature integration in memory.