@techreport{oai:ipsj.ixsq.nii.ac.jp:00218826,
 author = {石川, 裕太 and 向井, 宏太朗 and 中西, 功 and Yuta, Ishikawa and Kotaro, Mukai and Isao, Nakanishi},
 issue = {21},
 month = {Jul},
 note = {人に超音波を提示し，それにより誘発される脳波を用いて個人を識別する研究を行っている．多数の電極で同時に測定される脳波から周波数特徴と複数の非線形特徴を抽出し，それらを個別に学習機能を備えた識別器で判定した後，それらの結果を多数決判定することで 0%  の識別誤りを実現した．しかしながら，電極数 × 特徴数だけ識別器の学習が必要になるため，その削減が必要である．本論文では，まず特徴と電極をランダムに選択することでより少ない組み合わせが実現できることを見いだす．さらにその結果から特徴と電極位置を均等に選択する方針が導かれ，その考えが有効であることを見いだす．さらにその方針は関連性低い特徴を融合することにもつながることから，新しい特徴量として脳波の統計量を導入し，その考え方が正しいことを証明し，識別器の数を大きく削減できることを示す．, Person verification using evoked EEG by ultrasound has been studied. From brain waves simultaneously measured at many electrodes, a spectral feature and several nonlinear features are extracted as individual features. Using them, whether a user is genuine or not is verified by learning-based classifiers at each electrode. Final decision is performed by the majority vote using the verified results of all features at all electrodes. The verification error rate is reached to 0% but huge amount of time for learning many classifiers is required. In this paper, it is confirmed that randomly selecting electrodes used in features makes possible to further reduce the number of classifiers. The randomly-selecting is equivalent with evenly selecting electrodes for each feature and each electrode position. The effectiveness of the evenly-selecting is statistically confirmed. Furthermore, the evenly-selecting leads to fusing uncorrelated features. Thus, four statistical values of brain waves are introduced and the effectiveness of fusing uncorrelated features is confirmed.},
 title = {超音波による誘発脳波を用いた個人識別-特徴融合における関連性の検討-},
 year = {2022}
}