Abstract
Natural objects typically possess polychromatic surface textures. In cone-contrast space, the chromaticity distributions of natural objects form distinctive signatures which may enhance colour constancy and object recognition (Hurlbert et al., VSS 2007). Whether the human visual system exploits this information may depend on its ability to discriminate between polychromatic textures, and to integrate colour and texture information. Here we measure the discriminability of chromatic textures in a surface classification task.
Observers performed a 2AFC task, answering ‘yes’ or ‘no’ to: ‘Do the two patches arise from the same object's surface?’ Images of natural objects were obtained using a tristimulus-calibrated digital camera system under controlled illumination. Reference surface patches were taken from different locations on single objects. Alternative patches were created by rotating the reference chromaticity distribution from its starting angle in cone-contrast space around the origin, maintaining luminance values and the overall shape of the distribution. Two starting angles were used for each reference surface patch: ‘normal’ (the mean angle of the natural object texture) and ‘abnormal’ (rotated significantly away from ‘normal’). Stimuli were displayed on a calibrated CRT monitor.
Discrimination thresholds vary across objects but are generally below that necessary to detect changes in the chromaticity distribution caused by illumination changes. ANOVA revealed significant main effects of ‘normality’ and' chromaticity distribution size, as well as significant interaction effects between ‘normality’ and starting angle in cone-contrast space, as follows. Discrimination thresholds were significantly higher for ‘abnormal’ starting angles; e.g. discriminating between ‘blue’ kiwis is harder than for natural ‘green’ kiwis. Thresholds were higher for smaller distribution ranges. For ‘normal’ starting angles only, thresholds varied systematically with angle, being lowest for ‘yellowish’ hues.
The results suggest that texture is linked to colour in object representation, and that the human visual system is able to exploit polychromaticity of familiar objects in recognition tasks.
Funded by EPSRC (UK) (grant EP/D0687381/1).