Abstract
Object identities appear stable despite their constantly changing image properties. This may reflect a Continuity Field, a mechanism of perceptual stability built on serial dependence, in which objects appear more similar over time than they actually are. Although explored in psychophysics, the physiological basis of perceptual serial dependence and the Continuity Field are unknown. Here, we used fMRI to test whether responses in early visual areas are biased toward previous stimuli in an attention-dependent manner. Subjects viewed natural movie segments (fixating) while performing three different tasks in separate runs. The three tasks were free attending, during which the subject reported attending to the main character in the video, attending to the supporting (not main) characters, and attending to the background scene. We correlated the BOLD signal of one task with the BOLD signal of another task (used as a reference signal) shifted back from one to five TRs (TR = 900 ms). We computed correlation coefficients of the BOLD signals of the tasks with their respective reference signals. These coefficients quantify the serial dependence effect and the autocorrelation of the stimuli. Taking the difference of two such correlation coefficients cancels out the autocorrelation and factors out the temporal blurring of the hemodynamic response, thus revealing the difference of the serial dependence effect between two tasks. Since perceptual serial dependence is modulated by attention, we expected more serial dependence in the BOLD signal for face specific regions during the attend-to-face tasks. Our results show this expected difference, and this difference was significantly higher for the attended face than for the non-attended faces. These results suggest that the activity of the voxels in early visual areas are biased toward previously attended stimuli in a spatially selective manner, hinting at a physiological basis for the continuity field and the resulting perceptual serial dependence.
Meeting abstract presented at VSS 2016