指紋影像分割在指紋認證系統中是一個重要的前處理過程，設計良好的指紋分割方法可以提升指紋辨識的辨識率。非按壓式指紋改善了在衛生及維護上的問題。然而因為谷和脊之間的對比較低以及拍攝時造成的動態模糊，造成非按壓式指紋在分割上會比一般按壓式指紋更加困難。因此本研究提出一個結合紋理分析，機率神經網路(Probability Neural Network，PNN)以及粒子群最佳化(Particle Swarm Optimization, PSO)的非按壓式指紋分割方法。本研究的影像分割流程可分為擷取紋理特徵，使用PNN分類器進行影像分類，將分類結果輸出為分割影像。最後以PSO 對PNN 的平滑參數進行最佳化得到最佳的分割效果。實驗結果顯示，本研究所提出的非按壓式指紋分割方法擁有較佳的分割性能，未來可將此一研究成果結合行動裝置以發展為行動指紋辨識平台。;Fingerprint image segmentation is an important preprocessing step in Auto Fingerprint Identification System. A well designed fingerprint segmentation method can improve the accuracy of the feature extraction. Touch-less fingerprint eliminates hygiene and maintenance problems. However, inadequate contrast between the valleys and ridges of a fingerprint results in motion blurs during image capture, complicating the separation process of touch-less fingerprint methods. Therefore, this study proposed a novel touch-less fingerprint separation method, in which texture analysis, a probability neural network (PNN), and particle swarm optimization (PSO) techniques were combined. During the separation process, textural features were captured as images and processed using PNN classifiers. The processed results were then output as separate images. Finally, PSO was applied to optimize the PNN smoothing parameters to improve the accuracy of the separation results. The experimental results verify that the proposed method attained excellent separation performance. Therefore, this method can be integrated into mobile fingerprinting applications for mobile devices.
