| 摘要: | 隨著生物辨識技術的發展趨於成熟，許多系統逐漸運用生物特徵來取代傳統密碼以達到認證功能；雖然此方案解決了密碼認證系統的問題(例如：易受暴力攻擊、密碼過長而難以記憶等)，另一方面，卻也不可避免造成了其他缺陷（例如：生物特徵的存續、不變性，以及辨識率等）。
虹膜辨識作為生物辨識技術的一環，具有虹膜特徵持久存在、高辨識率等優勢，另一方面，人臉辨識則得益於較低的設備成本，並保持一定程度的辨識率；因而兩者皆用於設置重要服務的認證系統。然而，生物辨識系統的準確率主要取決於高自由度的特徵模式，低解析度的生物資訊影像將會造成特徵細節的損失進而大幅地降低辨識率；因而傳統上採用成本較高的取樣設備以獲得高解析度的影像。鑒於近年物聯網與移動裝置的技術熱潮，將成本較高的影像取樣設備佈署於移動裝置上，勢必造成整體設備費用的高昂，若使用成本低廉的設備也未必能滿足各種基於影像解析度的需求，例如：感光元件的硬體條件、高解析度的影像造成的儲存負擔、傳輸高解析度影像所消耗的頻寬等。
有鑑於此，本研究欲提出生物資訊影像的超解析度技術，使得原始的低解析度輸入影像能轉變為超高解析度影像，並得益於清晰的細節資訊來大幅地提升整體系統的辨識效果，並間接地降低低端設備的建置成本。
;Since the biometric solution is able to avoid the security problems which the traditional method may be compromised, for example, the password may be compromised by brute force attack or phishing attack, many systems gradually use biometrics to replace traditional passwords to achieve authentication functions. However, other potential defects of biometric are inevitably caused, such as the immutability of physiological characteristics, and the lower recognition rate in unconditional environment.
As a part of biometric technology, iris recognition has the advantages of features persistency and high recognition rate. On the other hand, face recognition benefits from lower equipment costs and maintains a certain recognition rate. Therefore, both authentication systems are commonly used for authentication services in business or military access control. However, the accuracy of the biometric system mainly depends on the features of high-degree-of-freedom. The low-resolution image will cause the loss of feature details and greatly reduce the biometric recognition rate. Therefore, the traditional solution attempt to apply the high-cost sampling equipment to achieve higher resolution images.
With the technology of the Internet of Things (IoT) become mature, the number of the IoT device will be expected to grow to 50 billion around the world in 2030, and the various service is already prepared to deploy on it. According to the traditional solution of iris recognition, the deployment of expensive image acquisition sensors on mobile devices will inevitably result in high retail price. However, the low-cost equipments may not be able to meet various requirements based on image resolution, such as the additive photosensitive element to catch the depth of field, the additive storage to store the high-resolution image, and the high bandwidth to transmit the high-resolution image.
In order to deal with above mentioned issue, the super-resolution technology have been proposed to generate the super-resolution image of biometrics from the low-resolution input. Benefit from clear detailed information of output biometrics image, the overall recognition system will be greatly improved, and reduce the cost of building low-end equipment indirectly. |
