|题名: ||Analysis and research on biometric identification system|
|作者: ||胡家幸;Hu, Chia-Shing|
|关键词: ||指紋分離;重疊指紋;臉部合成;Fingerprint separation;overlapped fingerprints;facial synthesis|
|上传时间: ||2015-05-04 10:02:13 (UTC+8)|
目前在人臉之不同年齡的合成系統中，都沒有強調五官對齊及扭曲影像的校正，若有這兩種情形，可能會導致影像上的失敗與合成上的不準確，在本研究中，我們提出一個整合ASM演算法與Log-Gabor wavelet的方法來達到人臉影像之老化/年輕化合成可逆系統，以便應用於失智老人之協尋。首先，我們利用ASM演算法可得到一組描述人臉五官特徵及輪廓的特徵集，將此組特徵集透過本系統的內眼角不變性及幾何不變性來達到人臉影像的校正。並再利用各特徵值間的相似程度，來判別臉型,以利搜尋與測試臉孔相似之樣本影像。接著，我們利用Log-Gabor wavelet轉換解析人臉影像之年齡紋理，以得到分解圖像，再過分解圖像數量的控制，有效地模擬出不同年齡之人臉合成，最後利用皺紋密度的方法來客觀判定合成的結果。
因此，本研究提出基於方向場之重疊指紋分離演算法。在方向場的估算部份，採用local Fourier analysis來決定方向場的方向，透過方向場的提供，再使用Gabor filter來取出正確的指紋。然而，錯誤的方向場會導致錯誤之指紋分離的結果，所以，在克服雜訊干擾的部分，利用機率密度函數的概念以及多尺度之技巧，來修正錯誤的方向場。並藉由相關性的量測，透過數學公式的計算，可以有效的鑑別分離之指紋的正確性。
The issues with missing persons in the present society and the lack of suitable channels to assist in locating those lost in the streets, particularly seniors with dementia whose memory degradations result in their inabilities to find the way home, are worrying as going home on their own is almost an impossible task. With their loved ones lost in the streets, the families could only search for the missing persons via the police, media and posting of photographs, during which all involved have to endure the anxieties, frustrations and helplessness of the process similar to finding a needle in a haystack. For the missing persons, their facial appearances may change in the years spent lost in the streets, hence by making their faces younger to facilitate better recognitions by those familiar with the missing persons and aid in the searches by the police or families, the opportunities for the seniors with dementia to return to their homes may be increased. Therefore, the development of the synthetic system for automatic aging/ reverse aging of facial models is not only an essential topic for the protection of seniors with dementia but also a significant contribution to the search efforts of the families.
The existing synthetic systems for the faces at various ages do not emphasize on the alignment of facial characteristics and the calibration of distorted images, which are conditions that may lead to failed attempts or inaccuracies in the synthesized images. In this study, a method integrating ASM algorithm and Log-Gabor wavelet is proposed to achieve a reversible synthetic system for the aging/reverse-aging of facial images, which may be applied to the searches for seniors with dementia. First, facial detection of the ASM algorithm are used to collect a set of features describing the characteristics and contours of the faces, which is then calibrated by the system via the invariance and geometric invariance of the inner corner of the eyes. The levels of similarity between the feature values are utilized to determine the face types for searching and testing with similar sample images. Then the Log-Gabor wavelet transformation is implemented to analyze the aging textures of the facial images to obtain the decomposed images, so that the synthetic faces of various ages may be effectively simulated by controlling the number of decomposed images and finally the wrinkle intensity method is applied to objectively determine the results of the synthesis.
Furthermore, in the dawning era of e-banking, e-commerce, smartcards, 3C products and cloud technologies, automatic personal identification has become an extremely important topic as the modern society places ever-increasing emphasis on the privacy and secure protection of personal information. Password identification is being phased out gradually due to its low levels of security. Therefore, the preference for the use of characteristics naturally inherent and unique to each human being as personal passwords for identification is now being widely applied in numerous types of products in countries around the world.
The “fingerprint” is unique, portable, difficult to forge, could not be forgotten and loaned, hence these properties render “fingerprint identification” as the top choice amongst the biometric identification methods at present.
Although the fingerprint identification technology has developed rapidly over the past 40 years, some challenging research topics remain to be resolved. The processing and matching of overlapping fingerprints, created when one or more fingers with multiple contacts on the same location of an object, is a challenging issue lacking attention. However, as the existing minutiae extraction algorithms assume only one fingerprint per image, the overlapping fingerprint data could not be properly processed. Therefore, the effective separation of overlapping fingerprints is an extremely important and essential process.
Hence, this study proposes an algorithm for the separation of overlapping fingerprints based on the orientation fields. The local Fourier analysis is utilized for the initial orientation field estimation and the Gabor filter is subsequently used with the orientation field to extract the fingerprint information corresponding to the orientations. However, the wrong orientation fields may lead to erroneous results in the separation of fingerprints, thus to overcome the noise interferences, the concepts of probability density function and multi-scale technique are implemented for corrections. And the accuracy of the separated fingerprints may be evaluated effectively by using the correlation measurements with mathematical calculations.