ISSN 1671-3710
CN 11-4766/R

Advances in Psychological Science ›› 2016, Vol. 24 ›› Issue (Suppl.): 57-.

Previous Articles     Next Articles

Visual Information Processing Mechanism Revealed by fMRI Data

Jinpeng Li; Zhaoxiang Zhang; Huiguang He   

  1. Research Center for Brain-inspired Intelligence, Chinese Academy of Sciences, No. 95 Zhongguancun East Road, Haidian Dist., Beijing, China, 100190
    University of Chinese Academy of Sciences
  • Online:2016-12-31 Published:2016-12-31


PURPOSE: The understanding of the information-processing mechanism of human visual system is a major cognitive neuroscience and physiology research topic. The insight of the structure and function of human visual cortex might also help us explain the state-of-art performances of recent deep neural networks.
METHODS: The human subject looked at a sequence of natural images whose content includes animals, buildings, food, humans, indoor scenes, manmade objects, outdoor scenes and textures. We analyzed the functional Magnetic Resonance Imaging (fMRI) data of both the ventral pathway (V1-V2-V4-LO) and the dorsal pathway (V1-V2-V3-V3a/b) on the visual cortex, and performed PCA-SVM on the basis of voxel activities of di?erent regions to conclude the trend of classi?cation accuracy. The canonical correlation analysis (CCA) was also used to estimate the linear correlationship among considered regions.
RESULTS: We found that the classification performance improved hierarchically from lower-level regions to higher-level regions in both pathways, among which the LO, V3a and V3b BOLD signals were good classification basis no worse than the widely-used features such as GIST, HOG and LBP. Simultaneously, the performances of V3a and V3b voxels were very close to those of LO voxels. CCA results also showed relatively strong linear correlationship between V3a/b and V4/LO.
CONCLUSIONS: Primarily, the visual cortex disposes visual information hierarchically. When we tracked the information stream in the visual system, the representations of the stimulus grew more and more abstract and global for recognition, so the layer-stacked deep neural networks were better models for visual cortex than traditional shallow ones. Moreover, V3a and V3b located at the dorsal pathway were also involved in object recognition, which was further confirmed by CCA results. Consequently, when it comes to object recognition, we should consider the coordination mechanism between the two pathways rather than focusing on the ventral pathway alone.

Key words: visual cortex, ventral pathway, dorsal pathway, fMRI, classification