摘要
近年來隨著電子科技的快速發展，電腦運算處理速度不斷的提升，對於需要大量運算需求的生物認證技術而言，開始逐漸受到廣泛的重視，其中以指紋辨識技術的發展已達實用階段，不論在比對速度、辨識精確度與成本考量上都有相當不錯的成果，但仍未盡完善。在指紋辨識系統中，由於一般傳統的指紋特徵擷取演算法需先經過二值化與細線化處理，因此容易受到指紋影像雜訊影響而降低其特徵擷取的準確性，且有處理時間較久的缺點，所以本文乃採用追蹤紋線走向的方式直接在灰階影像上擷取指紋細部特徵點，以提升特徵點擷取速度與準確度。 本文主要針對熱感式灰階指紋影像之特性來研究並建立特徵擷取演算法的架構。首先在影像中以指紋邊界輪廓搜尋界定出指紋紋路區域範圍來節省特徵擷取演算法的運算量；再根據搜尋紋線追蹤起始點之位置來開始追蹤紋線並達到擷取指紋細部特徵點之目的；最後再藉由錯誤特徵點移除之步驟保留正確性較高的特徵點以增加特徵擷取準確度。 在實驗結果與討論中，本文對紋線方向角與特徵點擷取準確度均加以統計與分析，並深入探討變動追蹤跨越間距參數時對演算法擷取特徵點正確性與運算處理時間的影響。由實驗結果可知本文所提出之特徵擷取演算法架構對熱感式灰階指紋影像而言，當追蹤跨越間距為3個像素寬時，不論在執行速度或特徵擷取準確度上皆有較令人滿意的結果，其特徵點擷取正確率大約可達75%左右。
並列摘要
With the rapid development of electronics technology in recent years, the computing speed increases significantly. The computationally demanding biometrics techniques are more and more feasible and gradually receive great attention. The development of fingerprint identification technique has arrived at a practical stage, where computing speed, recognition accuracy, and even the cost are acceptable. However, they are not perfect. The traditional minutiae extraction algorithms in a fingerprint identification system require binarization and thinning as a preprocess. As a result, the minutiae extraction accuracy may be reduced under the influence of image noise and longer processing time is expected. Therefore this thesis proposes an approach to extracting minutiae directly from gray scale fingerprint images by a ridge tracing method in order to increase the minutiae extraction speed and accuracy. This thesis focuses its study on the development of minutiae extraction algorithms based on the characteristics of thermal fingerprint images. First, an effective fingerprint zone in an image is identified by a boundary searching method to reduce the computational load of the minutiae extraction algorithms. Then minutiae are extracted by tracing ridges from the starting points on the corresponding ridges. Finally, the minutiae extraction accuracy is enhanced by eliminating the false minutiae and preserving the ones with higher probability being correct. In the experimental results and discussion, this thesis analyzes the accuracy of ridge orientation and minutiae extraction. The accuracy of minutiae extraction and the required processing time for different tracing step sizes are also obtained and discussed. The experimental results show that when the tracing step size is 3 pixels, both the processing time and the minutiae extraction accuracy are satisfactory. In this case, the minutiae extraction accuracy is about 75% for thermal fingerprint images.