【作者】 付景广；

【导师】 王裕国；

【作者基本信息】 中国科学院研究生院（软件研究所） ， 计算机应用技术， 2003， 博士

【摘要】 尽管指纹识别的研究和开发已取得重要进展，但是指纹识别的应用在目前并没有获得普及，这是因为指纹识别在识别准确性和识别速度方面还远远不能满足很多实际应用的要求。进一步提高指纹识别的性能无论在理论上还是在应用上都具有十分重要的意义。鉴于此，本文综合利用数字图像处理、模式识别、计算智能等方面的知识，对指纹图像增强问题、指纹细节点提取与验证问题、指纹细节点匹配问题和指纹分类问题进行了有益的探讨，并取得了以下研究成果： (1)首次提出了基于二值图像的细节点提取方法。用行程匹配方法提取出代表没有分支的局部指纹纹线段的图段，根据图段的结构形式和图段之间的连接关系用自定义的规则进行细节点判定。由于无需进行图象细化处理，因此处理速度较快，且可以避免由于细化畸变而产生很多虚假细节点。实验表明了其有效性，为细节点提取提供了一种新的途径。 (2)提出了一种新的基于模糊几何特征和纹理特征的细节点验证方法。取细节点在原始灰度图象上的局部邻域，分析邻域中的模糊几何特征和纹理特征，以这些特征作为MLP神经网络的输入，实现细节点的真假验证。实验表明其效果比直接用局部邻域中象素点进行分类验证的方法要好。 (3)提出了综合利用遗传算法和模糊逻辑的细节点匹配方法。用改进的基于遗传算法的点匹配算法求出使对应点数目最多、匹配误差最小的细节点对应关系，根据得到的对应点数目和匹配误差大小利用自定义的模糊逻辑规则推理出匹配分值。该方法模拟了人类进行指纹匹配时的模糊性，匹配决策更加合理，实验表明具有合理的精度。 (4)给出并实现了一种基于Gabor滤波的指纹图象增强改进方法。求出每个象素点处的纹线方向和方向一致性，以及纹线的平均频率，用方向和平均频率调节Gabor滤波器对每个象素点进行自适应滤波，并利用方向一致性分割出不可恢复区域，实验表明具有很好的增强视觉效果。 (5)给出并实现了基于奇异点和指纹中心对称轴的指纹分类方法。如果能提取出足够的奇异点，则主要根据奇异点进行分类，否则利用指纹中心对称轴进行分类。实验表明具有较好的效果。 最后总结全文，分析了目前研究工作中需要进一步完善的地方，指出了今后工作的研究方向。更多还原

【Abstract】 While a significant progress has been made in the research and development of automatic fingerprint recognition, the application of the technology does not prevail at present. The reason is that the accuracy and speed of recognition is far from satisfactory to many practical circumstances. To improve the performance of automatic fingerprint recognition is meaningful both in theory aspect and promoting its application. For this reason, this thesis has discussed the problems of fingerprint image enhancement, fingerprint minutiae extraction and verification, fingerprint minutiae matching and fingerprint classification, using the knowledge of digital image processing, pattern recognition and computational intelligence. The major contributions of this thesis are listed in the following:(1) A minutiae extraction method based on binary image is proposed for the first time. Firstly the image segments are extracted from binary image based on run-length code matching where each segment represents a section of fingerprint ridge without bifurcation. Then the minutiae are determined by defined rules based on the structure of these segments and their link relations. Because no need to perform image thinning, the method is more faster and can avoid many spurious minutiae caused by thinning aberrance. Experimental results show its efficiency. The method provided a new way to fingerprint minutiae extraction.(2) A new minutiae verification method based on the fuzzy geometry features and texture features is proposed. Firstly take the local neighbor area of a minutia in the gray image. Then analyze the fuzzy geometry features and texture features in the local area and use these features as the input of a MLP neural network to realize the classification of true and spurious minutiae. Experimental results show that the proposed method is better than a method which using pixels in the local area directly.(3) A minutiae matching method combining genetic algorithm and fuzzy logic is proposed. Firstly using a modified points matching method based on genetic algorithm to find the best minutiae corresponding relations which having the most matching pairs and minimum matching error. Then reason the matching score based on defined fuzzylogic relations between the number of matching pairs, the matching error and the matching score. This method simulate the uncertainty of human decision-making when compare two fingerprint, so it’s more reasonable. Experimental results show that the method is accurate.(4) A fingerprint image enhancement method based on Gabor filter is presented and realized. This method estimates the orientation and orientation certainty at each pixel, computes the average ridge frequency. Then adjust the parameters of Gabor filter using the orientation and average ridge frequency at each pixel to realize adaptive filtering. The unrecoverable region is segmented based on orientation certainty. Experimental results show a very good visual enhancement effect.(5) A fingerprint classification method based on singular points and central symmetrical axis is presented and realized. If sufficient singular points can be extracted from fingerprint image, then classify fingerprint mainly based on singular points. Otherwise classify fingerprint based on central symmetrical axis. Experimental results show that the method is more efficiency.Finally, summarize the work of this thesis, analyze the improvements need to be done and give the directions of future work.更多还原