Abstract
Manufactured glossy surfaces can exhibit a wide range of appearance characteristics including haze, sheen, sparkle and goniochromaticity. Specifying, communicating, and controlling the appearance of glossy surfaces remains a challenge for industry as current measurement standards fail to capture the full range of appearances reliably. As a first step towards developing an industrial standard for glossy appearance, we probed perceptual gloss space using a set of 29 curved samples from the lacquer company "Lechler", which spanned a wide range of different appearances. Participants viewed the samples on a matte black background under a 6500K diffuse "White Screen" lamp and performed three tasks. All 32 participants first arranged the samples on the table according to their similarity, with multiple repetitions to organize the samples according to different aspects of their appearance, in descending order of the subjective saliency of the different characteristics. A camera and custom computer vision system recorded the spatial arrangements. Subsequently, 16 participants rated each surface for 22 verbally defined appearance characteristics (e.g., 'metallic', 'coated', 'iridescent', 'dark', 'smooth') using continuous sliders. The other 16 participants performed pair-wise similarity judgments for 16 of the samples. Perceptual distance matrices were derived from all three tasks and analysed using MDS. We also applied PCA to co-registered images of the samples under the same illuminant, to identify image cues associated with the perceptual similarities. We find a striking similarity in the arrangements from different subjects and tasks, which correlated surprisingly well with low-level image measurements derived the image analysis. This suggests a robust and systematic perceptual gloss space, at least when lighting and shape is standardized. The subjective dimensions were arranged hierarchically in 5 - 6 clusters, suggesting that a low-dimensional perceptual space captures most salient aspects of glossy appearance.
Meeting abstract presented at VSS 2017