使用光學式滾動指紋的影像拼接進行背景雜訊濾除以及基於紋脊方向性的濾波使影像清晰度提升之研究

Abstract

指紋識別系統 (fingerprint identification system)一直是生物識別系統(biometric system)中使用最廣泛的技術，並且可應用於行動裝置或可攜式裝置上，而如何避免誤判及增加辨識的效能一直是發展的重心。近年來，光學菱鏡薄型式 (FTIR with a sheet prism)指紋識別裝置開始應用於可攜式裝置，其具備便宜及灰階度高等優點，相對的是容易受識別環境的光線所影響形成背景雜訊，或於攜帶中容易刮傷取像鏡面而產生刮痕造成雜訊，而目前針對此雜訊處理的相關研究相當稀少且尚有提升之空間，且如何增強光學式指紋識別裝置的影像品質也是相當重要的研究。有鑑於此，本論文主要發展一種用於光學菱鏡薄型式指紋識別機之影像處理系統，利用提出的指紋影像增強(fingerprint image enhancement)演算法提高單張指紋影像的辨識率，其次利用指紋滾動(rolled fingerprint)過程中影像的前後關聯性，開發一個拼接程序的演算法，並結合指紋影像增強演算法成為一個完整的指紋識別系統，去克服取像時會遇到的外界環境干擾。 本論文所開發的新式演算法分別為利用巴特沃斯帶通濾波器(Butterworth bandpass filter, BBPF)提出一種指紋影像強化方法，可濾除高低頻率雜訊並設計出方向性濾波器(direction filter)於順著紋路方向增強正確的紋脊(ridge)資訊，可有效的隔絕紋路不清晰的部分，並能濾除雜訊。其次為開發一指紋影像拼接(fingerprint mosaicking)的技術，將指紋滾動中的前後影像差異值取出做接合，做一個動態指紋的取樣拼接，使結果能夠有效的濾除背景雜訊，以及非指紋本身的雜訊，並提供一個新的研究方向。實驗顯示本論文提出的指紋影像增強演算法於單張影像的識別率可提升14.4%，而指紋影像增強演算法結合影像拼接背景濾除演算法在不同的資料庫和環境下實驗，分別顯示未拼接及兩張影像與多張影像的拼接結果，提供將來在不同環境下指紋品質提升的參考。

Fingerprint identification system is widely used in biometric system, and furthermore it is full of potential to use in mobile devices. However, to enhance the recognition rate reliable and to keep off erroneous identification are the main issues in development. In recent years, FTIR with a sheet prism based fingerprint identification devices have been used in mobile devices due to their higher gray level and lower cost. It is easy to form noises from the background light or mirror scratches but the relative researches of improvement are rarely discussed. In this thesis, we developed a novel fingerprint image enhancement algorithm and integrated with fingerprint mosaicking process as a complete fingerprint identification system. This system has two characteristics, first, we presented an image enhancement algorithm, which used the Butterworth bandpass filter to percolate intermediate noises, and designed a direction filter to correct information of ridges that we can separate some unclear fingerprint lines and filter off noises effectively. The experiment results show the algorithms could improve the recognizable ability of single fingerprint identification, reach to 14.4%. Second, we used the fingerprint mosaicking techniques to characterize the difference value between the before and after images that we could mosaick the fingerprint images. Thus, the identification system which integrated with enhancement algorithm and fingerprint mosaicking technique shows the comparable results of single image and mosaicking dual images under different experiment surroundings and databases, respectively. These results could be good references to use in fingerprint identification enhancement.