隨著科技與網路的快速發展,有許多傳統圖書館結合資訊科技邁向圖書館數位化。但目前數位圖書館在認證使用者方面,大多以帳號密碼登入為主,可能有資訊安全上的疑慮。目前指靜脈辨識技術已在多個地方實際運用,如能把指靜脈辨識技術運用在登入數位圖書館上,將能提高閱覽時的安全性,又能增加便利性。目前在指靜脈辨識上大多是先將圖片預處理,凸顯特徵後再去做指靜脈辨識,過程繁瑣。因此本研究實驗是使用不經過預處理的圖像,讓深度學習模型辨識指靜脈圖像,藉此減少預處理過程。我們使用SDUMLA與FV-USM資料庫的指靜脈圖像資料做測試實驗,測試ImageNet LSVRC圖像分類大賽中較出名的深度學習模型。實驗結果比較不同模型的辨識度,最後以ResNet的辨識度最高。
With the rapid development of technology and Internet, many traditional libraries are moving towards digitization by integrating information technology. However presently most digital libraries rely on account and password log-in to authenticate users, thus there may be some concerns about information security. At present, finger vein identification technology has been applied in many fields. If this technology can be applied to access digital libraries, it will improve the security and convenience of reading. Currently, most features identified by digital vein identification is excuted after image preprocessing, which is a complicated process. Therefore, in this study, images without preprocessing were used to enable the deep learning model to identify the images of finger veins, thus reducing the preprocessing process. We used the digital vein image data from SDUMLA and FV-USM database to do test experiments to investigate the well-known deep learning model in ImageNet LSVRC image classification competition. The identifications of different models were compared among experimental results, and ResNet has the highest identification.