Abstract
Research on the contribution of different spatial bands to object recognition has been prolific, but there has been relatively little work examining how spatial factors and level of categorization task interact. In this study we examined the effects of spatial frequency filtering on object categorization at the Basic, Subordinate and Superordinate levels. In each trial, subjects were shown first a word at either the Basic level (e.g., Dog, Car, Boat, etc.), the Subordinate level (e.g., Collie, Limousine, Sailboat, etc.), or the Superordinate level (e.g., Animal, Vehicle) and then a picture of an object. Their task was to indicate if the pictured object matched the word. The picture could be either low-passed (50% cutoff at 8.0 cycles/image width), high-passed (50% cutoff at 16.0 cycles/image width) or full-bandwidth. The design was blocked along the spatial frequency and category level dimensions; trials were otherwise completely randomized. Reaction time and error rates were assessed. Our results show that while both superordinate and subordinate classifications are adversely affected by spatial filtering, basic-level categorization is robust to this manipulation. Subordinate classification proved especially vulnerable to low-passing. These findings are in agreement with previous studies suggesting that basic-level representations are robust to changes in image information (rotation, scrambling, etc.). These results also suggest that differences in the level of categorization task may explain in part differences in findings regarding which spatial bands are most effective for object recognition.
This work was supported by grants to CC from FCAR and to PM from NSERC and HFSP.