The cost of parallel processing in the human visual system

Abstract

Our environment is visually rich, containing a multitude of objects that can be defined by many different features, e.g. shape, colour, and motion. To navigate and interact with the environment, we must process this information efficiently. The human visual system can process information either serially or in parallel. While there is a clear timesaving benefit of parallel processing, its cost is less well understood. Consequently, the aim of this thesis is to address three key theoretical questions underlying the cost of parallel processing. The first aim was to determine how the capacity of parallel processing varies as a function of the detail of information extraction. Previous research has demonstrated that brief presentations of five and six motion signals can be differentiated; this suggests that up to five signals can be simultaneously processed. However, it is unclear how much information is being extracted, i.e. whether observers are extracting direction information from all five signals. To examine this we presented observers with multiple moving objects and evaluated their parallel processing capacity as a function of the information required to perform the task. We found that the resolution of parallel motion processing varies as a function of the information that is extracted; specifically, as information extraction becomes more detailed, the capacity to process multiple signals is reduced. The second aim was to investigate whether there is a cost to the fidelity of information that is processed in parallel. Previous research suggests that there may not be a cost associated with parallel consolidation of information from sensory to visual shortterm memory (VSTM). Here we examined this by first determining that motion direction, and possibly orientation, can be consolidated in parallel, then explicitly evaluating the cost to the fidelity of information consolidated in parallel, compared to serially. We found that there is a twofold cost associated with parallel consolidation: a reduction in resolution of encoded items due to spreading of spatial attention, and an increase in the likelihood of consolidation failure due to interference between items. The third aim was to examine whether the cost associated with parallel processing can ultimately explain its capacity. We extended our previous findings regarding the cost associated with parallel consolidation to examine whether the capacity of parallel consolidation results from biased competition, the same mechanism proposed to account for spatial attention and VSTM storage, as evidenced from the interference between items presented simultaneously. This was achieved by demonstrating that parallel consolidation performance is influenced by factors predicted by a biased competition model. Furthermore, we found evidence suggesting that the capacity may be as high as three, with increasingly poorer resolution and higher consolidation failure-rates. Together, these results demonstrate that a) parallel processing is limited by the complexity of information to be processed, b) there is a twofold cost of processing information in parallel, and c) that increasing the amount of information processed in parallel also increases this cost to the fidelity of the information and ultimately leads to the capacity of this process.