一個高精確的虹膜特徵擷取方法及其在虹膜辨識的應用

Abstract

本文提出一種新的高性能虹膜特徵擷取方法。首先我們規劃了一個簡單快速的瞳孔位置偵測方法接著以瞳孔中心座標為參考點，擷取環狀虹膜區塊影像，並將其切割多個等分的長條影像，我們應用索貝爾轉換去增強虹膜紋路特徵，再藉由垂直投影得到一維能量訊號，此一能量訊號經由一維小波轉換有效縮減其維度同時抑制其高頻雜訊。最後，基於此一虹膜特徵擷取方法，我們以機率式類神經網路分類器來進行虹膜識別的實驗。CASIA 虹膜資料庫被用去評估所提出的方法並與傳統方法的性能做比較。實驗的結果證明提出演算法在虹膜辨識不僅有相當好的辨識性能，同時有較小的特徵維度和辨識效率，十分適合實現於嵌入式系統或有硬體資源限制的即時系統。

In this paper, a novel technique is proposed for high performance iris feature extraction. First, we elaborate a simple and fast iris location method and extract an iris image by the center coordinate of pupil. The iris image will be stretched into a rectangle block and the block is segmented many parts. We adopt Sobel transform to enhance iris texture, vertical projection to obtain one dimension energy signal, and one dimension wavelet transform to reduce the feature vectors of the energy signal and restrain the high frequency noise. Finally, probabilistic neural network (PNN) is regarded as a classifier for iris recognition. A comparative experiment of existing methods for iris recognition is evaluated on CASIA iris image databases. The experimental results reveal the proposed algorithm provides superior performance in iris recognition, but it has still small feature dimension and recognition efficiency. These prove the proposed method is suitable for embedded system or real-time system in resource-constrained.