RICERCA ITALIANA     

Research program

Information integration in the visual system: Computational models of decision and sensory information processes

University Co-ordinator

Università degli Studi di PADOVA - PSICOLOGIA GENERALE - PADOVA(PD)

Research Unit Leader

Gianluca CAMPANA

Description

In the recent literature is still open the debate on how attention infuences elementary cognitive processes in vision. The hypotheses relative to the role of attention on modulation of integration of visual information comprise: an improvement of signal-to-noise ratio (tuning of neurons: Dosher & Lu, 2000), an increase of the signal through improvement of the perceptive representation of the stimulus (Yeshurun & Carrasco, 1998), a modulation of short and long range (inhibitory and excitatory) interactions between channels (Ito, Westheimer & Gilbert, 1998; Freeman, Sagi & Driver, 2001). Using stimuli defined by textures (stimuli made of repeated micro-structural elements of an image having constant spatial relationships), we want to investigate on one side the role of attention in modulating the integration of information on global perception of textures segmented from other textures (discrimination of different areas on a texture surface whose elements differ for a given simple feature such as orientation), and on the other side the role of attention in modulating the integration of information on perception of single elements embedded on a texture. For the two conditions to be comparable, stimuli should be identical in the two conditions, and made of texture elements all having the same orientation and spatial frequency, except a triplet of elements differing in orientation respect to the other (background) elements, which will segment from background elements as a texture bar. The triplet of "odd" elements will have a local (single element's) orientation either collinear or orthogonal to the global (bar) orientation. In two different experiments subjects will be asked to judge the global (bar) orientation, or the local (central element) orientation. It is important that, in all cases, (local) orientation contrast between odd elements and background elements is kept constant. Whereas in the first case information integration (grouping) of close elements is required for texture segmentation, in the second case integration of information betweeen elements should not be necessary for segmenting a single element from the texture array: orientation contrast per se should be sufficient. However, we know from the recent literature (Polat & Sagi, 1993; Ito & Gilbert, 1999; Gilbert et al., 2000; Freeman et al., 2001) that surrounding elements may modulate the response (at perceptual, neurophysiological and electrophysiological level) to a given target.
The question we can ask is whether the two types of integration are subserved or not by the same mechanissms, that is whether the two types of integration are sensitive to the manipulation of attention, configuration, or eccentricity.
It is our intention to verify whether, as with perceptual learning (textures: Casco, Campana, Grieco & Fuggetta, 2004); visual search: Ahissar & Hochstein, 2000), attention is able to modulate the response of those channels (selective for orientation and spatial frequency) that are relevant for the task, or whether attention simply increases stimulus spatial resolution therefore producing facilitatory or inhibitory effects depending from task, eccentricity or configurational factors (Yeshurun & Carrasco, 1998).
In particular, we want to verify:
a) whether the response to texture-defined stimuli or to single targets embedded on a texture varies depending on stimulus eccentricity;
b) whether there is an interaction between stimulus eccentricity, task (orientation discrimination of a texture which involves segmantation vs. orientation discrimination of a single target embedded on a texture) and configurational factors (flankers or texture elements collinear vs. orthogonal);
c) whether the effect of attention on configurational factors (collinear vs. orthogonal elements) is dependent on eccentricity (Yeshurun & Carrasco, 1998, 2000) in the two tasks (global texture vs. single target).
We aim to measure behavioural (accuracy) and electrophysiological (VEPs) correlates to the presentation of a group of texture elements (bar) made by a triplet of gabor elements that can be collinear or orthogonal to the global bar orientation. The use of gabor elements (sinusoidal gratings convolved on a gaussian) is particularly effective as they reproduce (and presumably stimulate) better than other stimuli, the response of simple cells with centre-periphery antagonism in the striate visual cortex. The use of gabors enables to stimulate selectively population of neurons with receptive field having a specific orientation and spatial frequency.
The stimulus will be made of a gabor matrix 9x9, where all gabors have the same spatial frequency. All gabors will have oblique orientation (45 degrees), except for the gabors that constitute the triplet that can have either horizontal or vertical orientation. The gabors in the triplet will be arranged so that they can have either collinear or orthogonal orientation to the orientation of the bar composed by the triplet. Stimuli will be presented very briefly (100 ms) and masked by a matrix of gabors with random orientation.
In the experiments we will measure behavioural (accuracy) and electrophysiological (VEPs) indices. Subjects will perform an orientation discrimination task on the texture bar during concomitant performance of a RSVP (rapid serial visual presentation) task. This type of paradigm will enable to manipulate in a parametric way the amount of attentive resources allocated to the bar orientation task. It is actually a dual task where the primary task (RSVP) consist in the presentation of either letters or numbers that rapidly switch (time of presentation is about 100 ms, with no interval) on the same spatial location. Numbers or letters have all the same colour except for one, the target. The primary task consist in RSVP target discrimination. Immediately after the RSVP task, it will be shown the secondary task of either globar or local orientation discrimination of texture bar. It is known (Raymond, Shapiro, and Arnell, 1992) that the more the target of the primary task –RSVP- is temporally close to the end of the sequence, that is temporally closer to the presentation of the stimulus of the secondary task, the more the attention available to perform the secondary task will decrease. This phenomenon has been termed "attentional blink".
The texture bar will be presented, in different experiments, in either foveal or more eccentric (5 degrees of eccentricity) location.
The important question we ask in this study is about the role of lateral interactions in tasks that do require grouping of single elements (texture segmentation) and in task that do not require grouping (decision on single targets) but where flanker elements can affect single target discrimination depending on contextual factors (collinearity) and depending on the effect of attention on contextual factors in both cases (textures
and single target). It seems that in literature there are no studies that compare the effect of attention on global texture segmentation (segregation and aggregation) with effect of attention on orientation contrast discrimination of a single element, and investigating effect of flankers in these two conditions using identical stimuli so that the effects can be compared. Besides, whilst most of the literature is focused on effects of increase of focal attention by cueing paradigms, we believe that we can obtain very relevant information on the role of lateral interactions between detectors when attention is gradually taken away from the orientation discrimination task by using a RSVP paradigm.
The experiments will include a dual task paradigm with a RSVP as a primary task and a texture bar (or central element) orientation discrimination as a secondary task.
The texture bar will be presented either foveally or more in periphery, at 5 degrees of ecccentricity. We will compare behavioural and electrophysiological data (in particular N2 and N3 components associated to texture segmentation) for the collinear and orthogonal conditions, separately for the different levels of attentional blink induced by the RSVP paradigm, and separately for the central and more eccentric presentation. We expect differences in performance depending on the different levels of attentional subtraction, especially in the difference between parallel and orthogonal configuration (interaction attention by configuration). This would indicate that the influence of atteention on feature integration depends upon contextual factors such as collinearity. This difference, however, could be different for the two tasks (local vs. global orientation discrimination), depending on eccentricity. If these difference exist, they could be revealed both with behavioural measures and with electrophysiological correlates (VEPs) recorded in occipital sites. The differerences in VEPs could concern not only latencies, but components as well: we expect a VEP amplitude correlate of the different levels of attentional subtraction in components with latency at around 160-200 ms in the global orientation discrimination task (as found by Caputo & Casco, 1999), whereas we expecct differences in earlier components in the local orientation discrimination task, since orientation contrast processing is believed to occur earlier and contestual influences such as collinearity could affect this process earlier with this task than in the global task (which involves texture segmentation and elements grouping).