Viewers display consistent attentional biases such as fixating the centre of the screen (Tatler 2007 JoV), or scanning a scene from left-to-right. Objects features are also known to contribute to attentional biases (e.g., Nuthmann & Henderson 2010 JoV); it is therefore likely that object position plays a role in such biases.
In the present object naming study, we characterize the effect of object location on overt attention and object naming frequency, comparing how different representations of an object's location predict naming. 24 participants named as many objects as possible after previewing photographic images (100 in total) for 5 seconds. This data was transcribed and a list of object label for each scene generated. All objects corresponding to these labels were annotated, resulting in an average of 29 polygons per image, with on average 14 distinct object labels.
To investigate the effect of position on attention we created "object maps" representing the distribution of fixated and named objects over our scenes. We find that the distribution of fixated objects can be fit by a distance metric x[sup]2[/sup] + ay[sup]2[/sup]. Setting a = 1.8 gives us R[sup]2[/sup] = 0.87. However, this model fails to account well for the distribution of named objects, R[sup]2[/sup] = 0.35. With the introduction of a vertical bias, bx[sup]2[/sup]+ ay[sup]2[/sup] + cy, the model successfully captures the distribution of named objects, R[sup]2[/sup] = 0.85.
This result suggests that the position of an object has different effects on overt attention and on how interesting, or memorable, an object is: while there is a strong bias to fixate objects in the centre of the screen, the distribution of named objects is shifted towards the lower half of the image.
Meeting abstract presented at VSS 2013