眉毛的外部特征提取及应用研究

发布时间：2018-05-10 09:12

本文选题：生物特征识别 + 眉毛　；参考：《安徽工业大学》2017年硕士论文

【摘要】：目前,人们普遍认为具有高度安全的身份识别手段就是利用生物本身所具有的独特特征(如指纹,人脸等)来进行识别。这种方式已经被广泛应用于金融服务、视频监控、信息安全、人机交互、刑侦判定、电子商务、出入境管理等众多行业范畴。鉴于人类的眉毛满足了利用生物特征进行识别而具有的唯一性、普遍性、稳定性以及易获取性的特点,所以,可以将眉毛作为身份识别的一种手段。眉毛具备显明的轮廓特征和纹理特征,但当前对眉毛的研究主要停留在单眉毛图像,且缺乏对眉毛的外部特征展开相关研究,因此仍然存在局限性,进一步展开关于眉毛的研究工作将会在生物特征识别方面具有重要的意义。本文开展的研究工作主要内容如下:(1)为了提取眉毛轮廓,本文首先简要的叙述了水平集算法的基本原理,然后重点研究了基于偏移场矫正的水平集模型(Li模型)。该模型不仅可以实现图像分割,也可以抑制图像的灰度不均匀性,但往往也因初始曲线设置不合理,进而导致演化时间过长。本文特别针对水平集初始曲线部分进行改进,其中引入基于伪球的边缘检测算子,再结合形态学闭操作和填充操作,将演化的初始曲线(初始轮廓)设定在靠近感兴趣区域(眉毛边缘),实现了对眉毛轮廓的粗定位,大大减少Li模型的水平集演化时间。(2)将基于伪球的边缘检测算子与Li模型进行结合,通过水平集演化获取眉毛图像中的眉毛轮廓,在此基础上,利用眉毛的几何特性,计算形状特征和方向特征,然后利用灰度共生矩阵法(GLCM)计算眉毛纹理特征,以特征向量方式构建眉毛的一种外部特征模型。(3)实验结果表明,在相同的迭代次数下,对比Li模型,本文方法得到的眉毛轮廓更准确;针对自建的自然眉毛图像库(100人),本文模型其身份验证的匹配率最高达90.59%,眉毛外部特征模型的单眉毛识别率可达86.1%,与HMM和2DPCA结果相当,双眉毛识别率略有提高(90.2%),针对没有浓淡区别的眉毛库,仅靠形状和方向特征模型,单、双眉毛识别率为88.1%和88.7%,说明也能达到识别眉毛的作用。
[Abstract]:At present, it is generally believed that a highly secure identification means is to use the unique characteristics of biology (such as fingerprints, faces, etc.) to identify. This method has been widely used in financial services, video surveillance, information security, human-computer interaction, criminal detection, e-commerce, immigration management and many other industries. Since human eyebrows satisfy the unique, universal, stable and easily accessible characteristics of biometric recognition, eyebrow can be used as a means of identity recognition. Eyebrow has obvious contour features and texture features, but the current research on eyebrow is mainly focused on single eyebrow image, and lack of related research on the external features of eyebrow, so there are still limitations. Further research on eyebrows will be of great significance in biometric recognition. The main work of this paper is as follows: (1) in order to extract the contour of eyebrow, the basic principle of level set algorithm is briefly described in this paper, and then the level set model based on offset correction is studied emphatically. The model can not only achieve image segmentation, but also suppress the inhomogeneity of image grayscale. However, the initial curve setting is unreasonable and the evolution time is too long. In this paper, the initial curve of the level set is improved, in which the pseudo-sphere based edge detection operator is introduced, and then the morphological closed operation and the filling operation are combined. The initial curve (initial contour) of evolution is set near the region of interest (eyebrow edge) to achieve the coarse location of the contour of the eyebrow. The level set evolution time of Li model is greatly reduced.) the edge detection operator based on pseudo sphere is combined with Li model, and the contour of eyebrow in eyebrow image is obtained by level set evolution. On this basis, the geometric characteristics of eyebrow are utilized. The shape and direction features are calculated, and then the grayscale co-occurrence matrix method (GLCM) is used to calculate the eyebrow texture features, and an external feature model of eyebrow is constructed by eigenvector. The experimental results show that, under the same iteration times, the Li model is compared. According to the self-built natural eyebrow image database, the matching rate of this model is 90.59, and the single eyebrow recognition rate of the external feature model of eyebrow can reach 86.1, which is similar to that of HMM and 2DPCA. The rate of double eyebrow recognition was slightly increased by 90.2%. The recognition rate of single and double eyebrows was 88. 1% and 88. 7%, which indicated that the eyebrow recognition rate could also be achieved by the shape and direction feature model.
【学位授予单位】：安徽工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】