Functional Research Of Scene-Selective Regions In Visual Decoding Application

The scene-selective regions in the brain play an important role in the navigational visual environment.Although different types of natural scene images usually have similar image priorities,humans can effectively distinguish between different categories of natural scenes.The research is mainly to analyze the functional role of several scene-selective regions.To explore the relationship between the neural response of the scene-selective regions and the low-level visual features as well as the semantic categories,the article uses multi-voxel pattern analysis and representation similarity analysis to decode scene images.Multi-voxel pattern analysis is mainly used to analyse the ability of PPA,RSC,OPA,LOC and V1 brain regions to distinguish different types of scene images or different filtered scene images.It is found that these brain regions can effectively implement the classification of scene category in high-pass and full-pass scene pictures,some brain regions have relatively poor classification of scene images processed by low-pass filtering.It is also found that two relatively low-level visual brain regions,LOC and V1,are more sensitive to distinguish low-pass images comparing to high-pass images.In order to further obtain the correlation between the activation of these brain regions and semantic categories as well as the low-level image attributes,the paper also used the representation similarity analysis.The results showed that the brain area activation of PPA,RSC and OPA is more related to the category and the V1 is closer to the frequency.Although the LOC brain area is more relevant to the category,it also shows a strong correlation with the low-level image attributes.In the process of representation similarity analysis,correlation analysis is performed not only on the visual characteristics of the stimulating image but the neural response of the scene selection region.In order to analyze the correlation between the scene selective brain region and the convolutional neural network,several classical CNN networks were also correlated with these brain regions,and the trends in the correlation between the CNN network and the neural response of the scene-selective region were found.Based on the resulting image,it can be seen that although the structure and depth of these CNN networks are different,the relationship between the changes with the activation of the scene-selective brain regions is basically similar.PPA,RSC,OPA,and LOC,which are more related to the scene category,show a correlation tendency of decreasing first and then increasing with the layers selected by CNN network,while V1 which is more related to low-level image attributes shows a trend of rising first and then falling.To some extent,it can be explained that the CNN network is closer to the processing of scene images in the V1 brain region in the previous layers,but closer to the PPA,RSC,OPA,and LOC in the later layers for the processing of scene images.