【作者】 张震；

【导师】 史小平；

【作者基本信息】 哈尔滨工业大学 ， 控制工程（专业学位）， 2017， 硕士

【摘要】 虹膜识别技术是人体生物识别技术的一种,因为其良好的性质,受到当今人们的广泛关注,是最具发展潜力的生物认证技术。图像超分辨率重构技术能够利用一组低分辨率图像产生高分辨率图像,可以保证清晰度的同时放大图像,该技术在医疗等领域得到广泛应用。当被识别者与人眼采集设备距离较远时,很难采集到足够清晰的虹膜图像,并因此影响虹膜识别系统的检测效果。为了解决这一问题,本课题采用了多图超分辨率重构技术,模拟人眼的连续拍摄情景,对多张低分辨率人眼图像进行重构达到提高辨识正确率的效果。首先,本课题实现了一个稳定可靠的虹膜识别系统,系统中结合二值图像边缘提取和Daugman算法两种算法,提高了虹膜定位的正确率,并实现了一维gabor,二维gabor和二维haar小波变换三种编码方式。其次,编程实现了双线性插值,自适应归一化卷积,迭代反投影和凸集投影四种超分辨率重构算法,并比较了四种算法的运行速度和重构效果。然后,根据人眼图像的采样间隔和被测者的配合程度,模拟了两种应用场景,在时间间隔短且被测者配合的情况下又分为先对人眼图像重构再虹膜定位和先分别虹膜定位再对虹膜图像重构两种应用,并对比了辨识结果。最终,确定了双线性插值和凸集投影算法比较适用于虹膜识别的结论,验证了图像超分辨率重构在虹膜识别中具有应用价值。更多还原

【Abstract】 Human iris recognition technology is a kind of biological recognition technology,because of its good properties,received extensive attention of the modern people,is one of the most promising biometric authentication technology.Image super-resolution reconstruction technique can make use of a set of low-resolution images to produce high-resolution images.The technology can ensure the image definition of magnified images at the same time,which is widely applied in medical and other fields.In general,there is always an inevitable long distant between collection devices and human eyes,it is difficult to collect clear enough images,and thus affect the detection effect of the iris recognition system.In order to solve this problem,this subject adopts multiple super-resolution reconstruction technology,simulate the continuous shooting scene of human eyes,and reconstruct the low-resolution human eye picture to a high-resolution one to improve the effect of the recognition accuracy.First of all,this subject build a stable and reliable iris recognition system,which combines binary image edge extraction and Daugman algorithm to improve the iris localization accuracy,at the same time,Implements 1D-gabor,2D-gabor and 2D-haar wavelet three kind of encoding method to choose from.Second,this subject implements the Bilinear Interpolation,Adaptive Normalized Convolution,Iterated back-projection and Projections onto Convex Sets four kinds of super-resolution reconstruction algorithm,and compare them from the angle of the running speed and the reconstruction effect.Then,according to the human eye image sampling interval and the degree of cooperation,this subject simulate the two kinds of application scenarios.In the case of time interval is short,meanwhile,the subjects is cooperate enough,it can be divided into first reconstruct human eye images or first locate iris,respectively,and then contrast the identification result.In the end,this subject had the conclusion of the Bilinear Interpolation and Projections onto Convex Sets algorithm is more suitable for the iris recognition,verified that image super-resolution reconstruction has its application value in the iris recognition.更多还原