Jun Saiki Graduate School of Human and Environmental Studies, Kyoto University

Long-term memory of visual shitsukan cannot be reducible to perceptual and kansei features of images. Also, shitsukan long-term memory is known to have biases. Effective visual communication through image thus needs quantitative understanding of interaction between image memorability and memory biases. This study aims at clarifying characteristics of shitsukan expression both memorable and bias-free using image transfer technique and continuous report paradigm of visual long-term memory. We examine image memorability and bias and their interaction by combining image memorability estimation using DNN, quantitative image transfer technique with image generative AI, continuous report paradigm, which is a standard experimental paradigm for visual long-term memory, and eye-tracking data reflecting perceptual, emotional, and semantic information of images. This study tries to establish theoretical bases for constructing memory support system to generate memorable and bias-free images in the future.