一種新的手指紋大分類的方法

Abstract

本論文提出一個新的手指紋大分類方法, 將指紋區分為Right loop, Leftloop, Whorl 和 Arch. 我們利用指紋紋路的整體流向來做分 類的工作. 本方法首先使用一種局部Thresholding方法, 將指紋影像二元 化, 同時將指紋的背景去除. 再來將二元化的影像細線化, 使得原來的影 像每一紋路的寬度都為一. 從細線化的影像中我們將整枚指紋分成6*5的 區塊, 並抽取每一區塊中指紋之主要走向, 再以方向矩陣表示此枚指紋的 形狀. 此部份主要是先將五條垂直線差入細線化的指紋影像中, 每條直線 會和指紋細線化後的線相交, 計算出每一交差點的斜率, 並將它量化成四 個方向, 然後將每一垂直線的所有交點切成六個部份, 找出每一部份中佔 最多同方向者, 使用此一方向代表此一區段, 如此即可得一6*5的方向矩 陣, 利用此一方向矩陣, 我們可以抽出一些重要的特徵, 這些特徵可以代 表指紋紋路Global的走向, 根據這些特徵, 還有這些特徵的相對位置, 我 們將指紋分類. 實驗結果顯示本系統較不受雜訊的影響, 具有很高的辨識 率. A new fingerprint classification method is proposed in this thesis. We classify fingerprints based on the global ridge shapes. The fingerprint image is first locally thresholded and the background is removed. After thresholding, we use the line following technique to obtain a thinned image. Then we establish a 6*5 directional matrix, which represents the global ridge shapes of the fingerprint. In order to establish the directional matrix, we insert five vertical lines into the thinned image and for each intersection point between one of the vertical lines and the thinned fingerprint, we calculate its ridge slope. Then we quantize these slopes into four directions using nonuniform quantization. After all directions of the intersection points are found, the intersection points in each line are divided into six parts. In each part, we calculate the number of each appearing direction and find the direction with maximum number. Then a directional matrix of 6*5 is generated, and some features are extracted from the directional matrix to represent the global ridge shapes. We finally classify fingerprints based on the combinations of these features. Experimental results are given to show that the system has a high classification rate.