Most of the time we are not passively viewing scenes but want to extract behaviourally relevant information. In addition, objects do not often occur in isolation outside the visual scientist’s laboratory but are embedded in complex visual scenes. If the brain is to be adaptive, it needs to process visual information with regards to its context. Thus perception is not purely determined by the specific input to the retina but depends on the surrounding scene, objects, attention, memory, prior knowledge, expectations and predictions. Traditionally, the visual system in the human brain has been viewed as having a hierarchical organisation with signals travelling in one direction: input from the eyes arrives at "lower" order areas, which then transmit their computations to "higher" order areas. As one moves up the hierarchy, visual areas code more complex and more abstract information, and after the final processing stage, the system gives an output. However, in reality things are not so simple. In fact, in the primary visual cortex (V1), which is one of the first visual processing stages in the brain, external stimuli constitute less than 10% of the total input. The rest of the input originates from internal connections, either within V1 itself or via signals arriving from "higher" areas, back down to V1. In this way, "higher" areas can tell "lower" ones about the bigger picture and the neighbouring elements. This internal processing in the brain is the mechanism which provides context and enriches the information reaching us from the external world. The signals arriving to V1 from the retina are referred to as feedforward, while the signals going in the opposite direction, from higher areas back to V1, are called feedback. Each neuron responds to its preferred stimulus in a specific region of the visual field, called the receptive field. Feedforward signals act on the central region of a neuron’s receptive field, while feedback signals act on a larger surround region and thus are able to inform the centre about the surrounding context. However, it is not well established which aspects of the surrounding scene define these contextual interactions. This thesis investigated the influence of the scene surround on feedback to V1. We aimed to establish how the scene surround contributes to informative feedback signals. An introduction about what is already known regarding the function of feedback and the information it transmits is provided in Chapter 1. I give an overview of the previous studies which highlight the various contextual roles of feedback, such as perceptual grouping, contour and object completion, expectation, attention and prediction, as well as being the mechanism allowing visual imagery. Chapter 2 aimed to address whether feedback provides coarse or fine-grained information about the surrounding scene. Since during normal viewing both feedback and feedforward signals are present, we investigated feedback signals in isolation by using a partial occlusion paradigm to remove meaningful feedforward input in a specific region of the scene. We filtered the scene surrounding the occluded region into a fine-grained and a coarse version. We also varied how much information was shared between the fine-grained and coarse version of the same scene. This was done to investigate whether the information feedback carried was tightly tuned to the spatial scale of the surrounding scene, or whether the information it contained was similar across the two types of the scene surround. We found that the feedback contained signals about both coarse and fine-grained surrounds, but there was also some overlap between these feedback signals. In addition, we found that the feedback information did not correspond to a direct "filling-in" of the missing feedforward input, suggesting that feedback and feedforward signals represent the scene in different ways. In Chapter 3 we took a closer look at the amount of meaningful scene surround that is necessary to elicit informative feedback signals. The results showed that increasing the amount of scene information in the surround resulted in more meaningful feedback signals. We confirmed our earlier finding that the feedback information in the occluded region is dissimilar to the corresponding feedforward input when the feedforward region is isolated from the scene surround. Adding the scene surround to the feedforward stimulus increased this feedback/feedforward similarity. Overall, these findings point to the notion that feedback signals combine with feedforward input under normal visual processing. Isolated feedforward input in the absence of the surround provides V1 neurons with impoverished information. Neighbouring elements of the scene or its overall global structure can be sources of context. In Chapter 4 we explored which regions of the scene surround contribute the most to the contextual feedback signals arriving at V1 – is this limited to only local neighbouring regions or does the feedback directly contain information about the overall global image structure, taking into account distant retinotopic regions as well? In the first experiment, we used simple global structures made up of four Gabor elements and showed that such simplistic shapes failed to induce contextual feedback into the occluded region. However, in the presence of feedforward information, we saw that feedback from the local surround combined with identical feedforward input to give rise to different activity patterns in that feedforward region. This suggests that feedback may be recruited differentially depending on whether feedforward stimulation is present or absent. In the second experiment, we used natural scenes and tested whether contextual feedback can originate from a distant retinotopic region in the situation when the local scene surround was not informative. We manipulated scene information in a distant retinotopic region (in the opposite hemisphere) while keeping the local neighbouring surround information the same. The results showed a lack of meaningful feedback in the occluded region, and that feedback from the distant surround had a negligible effect on the identical feedforward information, in contrast to the finding obtained previously with the local surround. These findings suggest that feedback preferentially originates from nearby regions and provides context to disambiguate local feedforward elements. Therefore context about the global scene structure may arise from a series of local surround interactions. Chapter 5 summarises these findings and discusses the overarching themes regarding the content of feedback and its role in full visual processing. At the end, I propose some future research directions.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:712630 |
| Date | January 2017 |
| Creators | Revina, Yulia |
| Publisher | University of Glasgow |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | http://theses.gla.ac.uk/8016/ |
Page generated in 0.2433 seconds