基于模块化神经网络的人耳识别

发布时间：2018-06-06 18:42

本文选题：人耳识别 + 模块化神经网络　；参考：《辽宁科技大学》2016年硕士论文

【摘要】：生物特征识别是利用人类特有的生理或行为特征对个人身份进行识别的技术。它提供了一种高可靠性、高稳定性的身份鉴别途径。而人耳识别是这一领域的一个新的分支,同时也是模式识别和计算机视觉极具挑战性的课题之一。与其他生物识别技术相比,人耳有其独特的生理和结构优势,使得它在公共安全、信息安全等领域有着潜在的应用前景,越来越受到人们的关注。现有的利用人耳特征进行的身份识别方法大部分需要进行图像预处理或复杂的特征提取,还有分类器的选择。而选择哪种特征和分类器进行识别,对最终识别率有很大的影响。神经网络技术是在模式识别中的成功应用。相对于传统的神经网络算法,模块化神经网络在模式识别中对于参数的选择、新增样本学习方面有很强的提高。本文针对模块化神经网络在人耳识别上的应用进行了详细的研究,主要有以下几方面工作:(1)本文对人耳图像库中的所有图像进行了预处理。运用中值滤波进行平滑处理,双线性差值进行尺寸归一化处理,灰度直方图均衡化进行去光照影响,使处理后的人耳图像更适合人耳识别研究。(2)对主成分分析理论进行了学习和研究,并用其进行了人耳图像的特征提取。(3)对神经网络算法进行了学习和研究,阐释了“模块化”这一概念,提出模块化神经网络的结构设计、模块化神经网络的优势。讨论了模块化神经网络应用在人耳识别上的可行性,并分析了识别出错原因和改进方向。(4)为了进一步提高识别的精度,提出分层遗传算法(HGA)对模块化神经网络的优化。通过优化模块化神经网络的结构和设计参数来提高人耳识别率。把训练好的网络进行人耳识别实验,通过对比实验验证了该方法的提高和可行性。
[Abstract]:Biometric identification is a technique to identify individual identity by using human physiological or behavioral characteristics. It provides a high reliability, high stability identification approach. Ear recognition is a new branch of this field, and it is also one of the most challenging topics in pattern recognition and computer vision. Compared with other biometrics, human ear has its unique physiological and structural advantages, which makes it have a potential application prospect in public security, information security and other fields, and has attracted more and more attention. Most of the existing identification methods based on ear features require image preprocessing or complex feature extraction, as well as the selection of classifiers. Which features and classifiers are selected for recognition have a great influence on the final recognition rate. Neural network technology is a successful application in pattern recognition. Compared with the traditional neural network algorithm, modular neural network in pattern recognition for the selection of parameters, the new sample learning has a strong improvement. In this paper, the application of modular neural network in ear recognition is studied in detail. The main work is as follows: 1) in this paper, we preprocess all the images in the ear image database. The median filter is used for smoothing, the bilinear difference is normalized for dimension, and the gray histogram equalization is used to remove the influence of illumination. Making the processed ear image more suitable for ear recognition research. (2) the principal component analysis (PCA) theory is studied, and the feature extraction of human ear image is carried out. (3) the neural network algorithm is studied and studied. This paper explains the concept of modularization, and puts forward the structural design of modular neural network and the advantages of modular neural network. This paper discusses the feasibility of the application of modular neural network in ear recognition, and analyzes the causes of recognition errors and the direction of improvement. In order to further improve the accuracy of recognition, a hierarchical genetic algorithm (HGA) is proposed to optimize the modular neural network. The recognition rate of human ear is improved by optimizing the structure and design parameters of modular neural network. The trained network is used in the ear recognition experiment, and the improvement and feasibility of the method are verified by contrast experiments.
【学位授予单位】：辽宁科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41;TP183

【相似文献】