在本文中提出一個新的手指靜脈血管影像特徵切割方法。手指靜脈血管紋路可做為生物辨識科技，因血管在體內不易改變及偽造，利用近紅外光照射手指產生的靜脈血管影像不僅包含幾何血管紋路，更含有不規則陰影及雜訊，由於在手指關節處肌肉群較少及光波動強度差異，會造成擷取時光線更易穿透而區域性過亮，所以我們必頇使用在亮度不均勻及雜訊干擾下仍有高穩定性的演算法。由於等階集合法具有強力的物件偵測及拓樸能力，而在傳統手指靜脈血管影像切割方法中，大多只使用二維濾波器處理，並無人研究使用等階集合法對血管影像做切割，因血管影像具亮度不均、低對比且高模糊度的複雜特性，容易產生錯誤切割結果，如使用以能量概念為基礎的Chan-Vese 模型時，亦因血管影像亮度不均，只切割出過亮的指節區域，所以本研究中探討如何使用並改良等階集合法，在疊代過程中增加背景影像不均勻修正演算，去除不均勻場以完成靜脈血管紋路的取得。從實驗結果中，我們可以有效的去除血管影像中亮度產生的不均勻場，並可同時得到不錯的修正影像及切割結果。 In this thesis, we propose a new approach to segment finger vein image. Finger vein pattern, embedded inside subjects‟ fingers, is difficult to be faked due to its nature of inter-individual characterization. The segmentation of finger vein pattern has drawn attention and has been used as a new biometric technology for personal identification. However, the inhomogeneous lighting or bone structure shadowing in the acquisition of finger vein image usually results in inhomogeneous shadowing patterns which can fail the image segmentation of level-set methods. Therefore, a reliable and robustness method is needed for image segmentation, even under inhomogeneous background condition. In contrast to traditional 2D filtering techniques, the proposed study adopted energy-based level set method to segment finger vein patterns, owing to the advantages of level set method in object detection and topology processing. By remodeling the Chan-Vese model with a background correction term, the inhomogeneous image background and finger vein pattern can be separated after iterative minimization of Chan-Vese energy function. Our experimental results showed that the modified Chan-Vese Level Set method is an efficient tool for image segmentation in bias-field contaminated image.
