Exposure duration influences holistic vs. part-based learning in visual object recognition
European Conference on Visual Perception
Hélène Devillez, John Rohrlich, David L. Sheinberg, Randall C. O’Reilly & Tim Curran
Embedded Files
The visual system is capable of simultaneously integrating different information to recognize objects. Usually, objects are made of different features. In this study, we wondered whether visual object recognition is a holistic processing or whether people have a tendency to represent and encode objects on a part-by-part basis. To do so, we used computer generated 3D species of objects made of 5 different features, with no diagnostic single feature. We trained participants to recognize species. They learned to distinguish between 5 species at a subordinate level whereas 5 other species remained untrained. We used a sequential matching task to test species recognition before and after training. Participants were either trained with long exposure durations (800 ms) and 3D viewing or were exposed to the stimuli for a short time (200 ms), to ensure a single fixation on the object. We hypothesized that subjects would demonstrate a more holistic encoding of objects with 1D short exposure durations compared to 3D long exposure. During long exposure condition, we found that training transferred to untrained species and inversion did not hurt performance. On the contrary, during short exposure condition, no transfer to untrained species was observed and participants performed better for upright compared to inverted trained stimuli after training. Learning appears to be part-based during the long exposure condition and more holistic during short exposure condition, suggesting that combinatorial learning (perfect generalization to novel combinations of same feature set) depends on separate fixations to the different features.
Page updated
Google Sites
Report abuse