Modeling object representations in natural vision

Papale, Paolo (2019) Modeling object representations in natural vision. Advisor: Pietrini, Prof. Pietro. pp. 116. [IMT PhD Thesis]

Text (Doctoral thesis) Papale_phdthesis.pdf - Published Version Available under License Creative Commons Attribution Non-commercial Share Alike. Download (9MB)

Abstract

Object perception relies on intensive processing in the human occitotemporal cortex (OTC). While its large-scale pattern of object selectivity has been widely described, looking at the computational processes controlling the spatial organization of OTC has proven challenging, since visual dimensions are mutually correlated (e.g., object shape and identity). In the present thesis, we investigated how different object properties that are relevant to behavior and share common variance are represented in our visual cortex in natural vision. In Chapter 1, we described how our exceptionally reliable visual system transforms continuous retinal signaling into meaningful objects. In addition, we demonstrated how this process is challenging and complex, given the unreliability of the retinal input and the widespread mutual correlations between behaviorally relevant object properties (e.g., shape and semantic category). In Chapter 2, we described an fMRI study on scene segmentation. In this first study, we analyzed brain responses during passive natural image viewing. Subjects attended to hundreds of natural scenes and we derived brain representations from each occipital region and compared them to parametric representations, so to reveal the inner filtering operated by each brain region. In contrast to strictly hierarchical and compartmentalized views on brain selectivity, the whole occipital lobe is involved in the high-level cognitive task of segmenting foreground and background, as early as V1. At the same time, contrast and spatial frequencies are represented also in higher visual regions such as V4 and LOC. However, due to the low temporal resolution of fMRI, the first study alone cannot resolve if those shared representations reflect a common spatiotemporal process. Thus, in a second MEG study, presented in Chapter 3, we derived brain representations in time and space of subjects attending to different objects. We compared these representations to modelderived representations comprising V1-like features, object shape and semantic category. We also employed a statistical approach that compute the relative weight (i.e., orthogonal component) of each model in explaining the MEG representations. By doing so, we found that a small cluster of posterior sensors independently processes all the tested features as early as 100-150ms after stimulus onset. Thus, the same features can be retrieved in the activity of multiple regions, and orthogonal components of those features are processed by the same cortical structures at the same latencies. Thus, is there a broader organization determining these observations? What is the link between coding of mutual and orthogonal object representations? In our third study, presented in Chapter 4, we employed fMRI to explore the spatial organization of sensitivity to mutual and orthogonal representations in the human visual cortex. Subjects attended to object pictures while performing an unrelated attentive task. By employing a variance partitioning method, we found that the weight of mutual representations increases along the visual hierarchy, from posterior to anterior regions. Overall, these results depict a complex picture of our visual cortex. First, there is not a clear selectivity hierarchy, but information spreads between regions: early cortical areas access to high-level representations and are involved in complex cognitive tasks (i.e., object segmentation). Second, concurrent processing of orthogonal object shape, contrast and category representations is fast (100-150ms) in the right posterior brain. And, third, the visual cortex encodes mutual relations between different features in a topographic fashion while object shape is encoded along different dimensions, each representing orthogonal features.

Item Type:	IMT PhD Thesis
Subjects:	R Medicine > RC Internal medicine > RC0321 Neuroscience. Biological psychiatry. Neuropsychiatry
PhD Course:	Cognitive, Computational and Social Neurosciences
Identification Number:	10.6092/imtlucca/e-theses/295
NBN Number:	urn:nbn:it:imtlucca-27317
Date Deposited:	27 Feb 2020 14:58
URI:	http://e-theses.imtlucca.it/id/eprint/295

Actions (login required, only for staff repository)

View Item