【作者】 张力；

【导师】 李大海； 吴庆富；

【作者基本信息】 四川大学 ， 光学工程（专业学位）， 2022， 硕士

【摘要】 信息时代的来临使得人们的生活高度信息化,伴随着大量信息的高速流通,随之而来的是日益严峻的网络与信息安全问题。图像作为一种基础的数据承载方式,其在信息行业的重要性不言而喻。在大数据、人工智能、5G等高新技术取得相当程度的进步后,以三维(Three-dimensional,3D)图像为代表的传输媒介将覆盖到社会的更多领域。3D图像视觉上层次分明且色彩鲜艳,具有很强的视觉冲击力,使得其在3D显示、医学成像以及人机交互领域具备很高的应用价值。除此之外,3D图像更加符合人眼的主观感知,使人有身临其境的感觉,在艺术欣赏、军事建模以及教学演示等领域潜力十足。种种因素使得3D图像成为了不可或缺的信息载体,在军事、医疗以及社会生活等各领域扮演着越来越重要的角色。在各种3D显示技术中,集成成像结构简单,无需相干光源,无需苛刻的光学环境,并且可实现真彩色和全视差的显示效果,有着非常广阔的研究和应用前景。在这种情况下如何保证基于集成成像的3D图像在网络上安全传输是一个不可忽视的问题。本文从集成成像原理角度,提出了两种3D图像光学加密算法,针对提出的光学加密算法是否足够安全,是否适合实际应用,进一步提出了基于多尺度条件对抗神经网络的3D图像加密分析算法,具体内容如下:第一,首先对集成成像原理和技术分类进行了介绍,并详细地阐述了基于集成成像原理的3D图像记录和再现过程。为了提高实验效果,我们引入了计算集成成像,通过光线反向跟踪渲染技术可为实验提供大量原始数据。此外还论述了基于细胞自动机的数字图像编码技术,最后概述了密码分析技术和深度学习的基本理论。第二,提出了一种基于集成成像和细胞自动机编码的3D图像加密算法,目前大多数加密算法都是针对2D图像设计的,而针对3D图像的加密过程非常复杂。因此我们提出了基于集成成像和细胞自动机编码的3D图像加密算法,通过光线反向渲染技术将3D图像记录在包含3D信息的微图像阵列中。然后通过细胞自动机生成的高质量随机序列用于对微图像阵列进行加密,加密过后的3D图像无任何明文特征。此外,仿真实验和光学再现实验验证了我们提出的加密算法具有密钥敏感性和鲁棒性。第三,介绍了一种基于集成成像和鬼成像的3D图像光学加密算法,由于目前3D图像的获取和加密是分开进行的,3D图像巨大的数据量会使图像处理和传输效率大大降低。为此我们提出一种基于集成成像和鬼成像的3D图像加密技术,可同时实现3D场景的获取与加密,通过控制采样率可对加密后的密文进行压缩,并为接收方提供不同清晰度的3D图像。第四,在分析了所提出的两种3D图像加密算法的特性后,为了进一步评估提出的基于集成成像的3D图像光学加密算法的安全性,我们提出了一种基于多尺度卷积核条件对抗神经网络的3D图像加密分析方法对基于集成成像和细胞自动机编码的3D图像加密算法进行安全性分析。在实验过程中,我们通过选择明文攻击的方式获得的一系列密文-明文对来训练我们提出的多尺度卷积核条件对抗神经网络。训练过的神经网络模型,在无密钥的前提下,可实现对3D密文图像的破译和光学重建。此外,所提出网络模型在2D图像密文分析领域同样适用,为基于集成成像的3D图像加密系统的安全性分析提供了新的研究方法。更多还原

【Abstract】 With the vigorous emergence of a new round of scientific and technological revolution,the network security problem becomes increasingly serious.As a basic data structure,image is very important in the information industry.With the advancement of the big data,artificial intelligence and 5G,three-dimensional(3D)images will be applied to more fields of society.Due to the strong visual impact of 3D images,it has high application value in 3D display,medical imaging and humancomputer interaction.In addition,3D images conform to the subjective perception of the human eye,which makes people feel immersive,and have great potential in art appreciation,military modeling,and teaching demonstrations.These factors make 3D images an indispensable information carrier,and play an increasingly important role in various fields such as military,medical and social life.Among various 3D display technologies,the integral imaging is simple,does not require coherent light source and harsh optical environment,can display colorful 3D images,which makes it has very broad research and application prospects.Under such background,how to ensure the security of 3D images based on integral imaging is a very significant issue.In this paper,from the perspective of integral imaging principle,two 3D image optical encryption algorithms are proposed.In order to verify the security of the proposed optical encryption algorithm,a cryptanalysis for light-field 3D cryptosystem based on multi-scale conditional adversarial neural network is further proposed.The research content is as follows:1.Firstly,we introduce the principle and technical classification of integral imaging,and describe the recording and reconstruction of 3D image.In order to improve the experimental effect,we introduce the computer-generated integral imaging,which can provide a large amount of raw data for the experiment through the ray reverse tracing rendering technology.In addition,we also discussed the digital image coding technology based on cellular automata,the basic theory of cryptanalysis technology and deep learning.2.A 3D image encryption algorithm based on integral imaging and cellular automata is proposed.Most of the current encryption algorithms are designed for 2D images,and the encryption process for 3D images is very complicated.Therefore,we propose a 3D image encryption algorithm based on integral imaging and cellular automata.The 3D images can be recorded in an element image array through ray inverse rendering technique.Then the high-quality random sequence generated by the cellular automaton is used to encrypt the element image array.Furthermore,simulation experiments and optical reconstruction experiments verify the key sensitivity and robustness of our proposed encryption algorithm.3.A 3D image optical encryption algorithm based on integral imaging and ghost imaging is proposed.Since the acquisition and encryption of 3D images are carried out separately,it will greatly reduce the efficiency of image processing and transmission.To solve this problem,we propose a 3D light field image encryption technology based on integral imaging and ghost imaging,which can simultaneously achieve 3D scene acquisition and encryption.By controlling the sampling rate,the encrypted ciphertext can be compressed,and we can provide the receiver with varying degrees of clarity 3D images.4.After analyzing the characteristics of the proposed 3D image encryption algorithms,in order to further evaluate the security of our proposed 3D image optical encryption algorithm based on integral imaging,we propose a conditional adversarial neural network based on multi-scale convolution kernels to attack the proposed 3D image encryption algorithm based on integral imaging and cellular automata.The experimental results prove that the trained neural network model can optically reconstruct 3D ciphertext images without a key.In addition,the proposed network model is also applicable in 2D image ciphertext analysis,which provides a new research method for the security cryptanalysis of 3D image encryption system based on integral imaging.The cryptanalysis we propose provides a new research method for the security analysis of 3D image encryption systems based on the integral imaging.更多还原