Abstract
Many natural objects are characterised not only by shape and mean colour, but also by their particular chromatic textures: the speckled red of a strawberry is distinctively different from the streaky red of an apple. While the roles of colour and shape have been well explored in object recognition, chromatic texture has not. Here we study the roles of mean colour and texture - and their interaction — in an object classification task using familiar objects.
Images of natural objects were captured using a tristimulus-calibrated digital camera under controlled illumination. Reference surface patches were taken from different locations on single objects, then manipulated by changing either their original texture or colour (e.g. combining banana texture with carrot colour). Stimuli were presented on a calibrated CRT monitor.
Observers performed a three-way speeded classification task for 3 stimulus sets: uniform colour patches (red, green or blue), whole natural object images (shape cue intact) and natural surface patches (‘congruent’, i.e., reference patches, and ‘incongruent’, i.e., manipulated patches). For the latter two sets, 3 groups were formed from combinations of 7 objects (example classification: potato, lime, or carrot). The task was performed at 2 different presentation times (40 ms and 250 ms). Observers were able to perform the task, even for incongruent patches. Classification performance (reaction time and accuracy) for whole object images was effectively the same as for uniform colour, at both presentation times. Classifications for incongruent patches were slower than for congruent patches, most pronouncedly at 40ms. Incongruent textures slowed down classification by colour more than incongruent colours impeded classification by texture, only at the shortest presentation.
The results strongly suggest that texture and colour interact in object recognition and, at least for this task, texture plays a more dominant role than mean colour in object classification.
This work was funded by the Engineering & Physical Sciences Research Council (UK) (grant EP/D0687381/1).