Investigation of Preprocess on Systematic Generation of Face Pareidolia using Cycle Consistent Adversarial Networks

カテゴリ: 部門大会

論文No: SS3-4

グループ名: 【C】2022年電気学会電子・情報・システム部門大会

発行日: 2022/08/24

タイトル(英語): Investigation of Preprocess on Systematic Generation of Face Pareidolia using Cycle Consistent Adversarial Networks

著者名: 遠藤 良峻（岩手大学）,下條 信輔（カリフォルニア工科大学）,張 潮（福井大学）,明石 卓也（岩手大学）

著者名(英語): Yoshitaka Endo (Iwate University),Shinsuke Shimojo (California Institute of Technology),Chao Zhang (University of Fukui),Takuya Akashi (Iwate University)

キーワード: CycleGAN|精神物理学|パレイドリアパレイドリア|CycleGAN|Psychophysics|Pareidolia

要約(日本語): The pareidolia, a psychological tendency, is perception of a specific object from others. There is currently no indicator represents relation between the pareidolia-inducing image and its pareidolia-inducing power to the best of our knowledge. Nevertheless, such an indicator is use-ful for studies in the neuroscience field or diagnosis for patients like the pareidolia test. Therefore, we have investigated the systematic genera-tion of the pareidolia-inducing image on the pareidolia-inducing power from face and natural scene data sets. In our previous study, we have found that using a general face dataset for face pareidolia generation yields in unnatural results because of some factors such as edge and background. In this paper we apply a series of preprocessing for the face data set and compares the generative results produced by CycleGAN with/without the preprocessing.