Taiki Orima^1,2 (), Hiroshi Ban^1,3; ¹Center for Information and Neural Networks (CiNet), National Institute of Information and Communications Technology (NICT), ²Japan Society for the Promotion of Science, ³Department of Frontier Biosciences, the University of Osaka

The human visual system is remarkably sensitive to the realness of images: we can correctly distinguish real photographs from paintings or computer-generated images even if they look realistic. This ability is vital for us to make correct judgments in our daily lives. However, the neural representation of realness remains unclear. Here, to investigate where and when in the brain the realness of images is processed, we conducted fMRI and MEG experiments in which participants (16/13 for fMRI/MEG) viewed real and unreal images (6,900 each). Decoding analysis showed that the real/unreal condition could be significantly classified from all the pre-defined regions of interest (early, ventral, and dorsal visual areas). However, the decoding results from higher dorsal visual areas could not be fully accounted for by image features (image statistics and style features) that differentiate real from unreal images. Additional decoding analysis from time-resolved MEG source signals revealed that decoding accuracy peaked at around 150 ms after stimulus onset in early and ventral visual areas. Interestingly, decoding accuracy in dorsal visual areas peaked at around 250 ms. These results suggest that the unknown aspect of realness, which is implicitly coded in the human brain, can be processed in higher dorsal visual areas. Next, to further investigate the neural and perceptual correlates of realness, we conducted an additional fMRI experiment using a parametric series of morphed images designed to modulate realness. Specifically, real and unreal images were morphed at 0/100, 10/90, …, 100/0 ratios. We recorded fMRI data from 9 observers while they viewed those morphed images. Subsequent decoding analysis showed that BOLD signals from higher dorsal visual areas notably classified perceived realness, suggesting that these areas specifically process realness. Our results indicate that realness of images is processed independently of image features in higher dorsal visual areas and at 250 ms.

Acknowledgements: Supported by CiNet, and by JSPS KAKENHI JP24KJ2222, JP25K172220.