@inproceedings{oai:ipsj.ixsq.nii.ac.jp:00201450, author = {大井, 健太 and 杉田, 誠 and Kenta, Ooi and Makoto, Sugita}, book = {コンピュータセキュリティシンポジウム2019論文集}, month = {Oct}, note = {筆跡鑑定の結果は裁判での重要な証拠となるため経験と知識を持つ専門家による鑑定が重要となる.しかし,現在の日本では筆跡鑑定人を名乗るための資格は必要とされていない.そのため筆跡鑑定人の信頼性に鑑定結果が大きく左右されてしまうという状況が起こっている.筆跡鑑定に有用な特徴の抽出を専門家の知識を使用せずに行うことができればよいと考えた.そこで,機械学習を用いて筆者識別実験を行い,抽出された特徴が筆跡鑑定に利用できないか調査した.本稿では,収集された筆跡を画像と捉え既に画像認識について有効性が確認されている畳み込みニューラルネットワーク(CNN)を用いて筆者識別実験を行った.CNNを用いたモデルによる実験で筆跡鑑定人の信頼性に左右されない特徴抽出が可能ということが分かった., Since the results of handwriting appraisal are important evidence in court judgment, it is important for experts with experience and knowledge to ex-pertise. However, in Japan it is not necessary to qualify for identifying a handwriting expert. As a result, the situation that the appraisal result is largely influenced by the reliability of the handwriting expert witness has occurred. I thought that it would be good if we could extract features useful for hand-writing experts without using expert knowledge. Therefore, we conducted an author identification experiment using machine learning and investigated whether extracted features can be used for handwriting expertise. In this pa-per, we conducted a writer identification experiment using a convolution neural network (CNN), which has already gathered handwriting as an image and has already been confirmed valid for image recognition. Experiments us-ing CNN model showed that feature extraction which is not influenced by re-liability of handwriting experts is possible.}, pages = {1116--1121}, publisher = {情報処理学会}, title = {機械学習を用いた筆跡鑑定}, volume = {2019}, year = {2019} }