Invariant representation of information and information quantity in image-processing tasks

A model of a signal (image) containing the information codes in its own digital virtual memory independent of stretching, packing according to intensity, and other prescribed image transformations is proposed. Virtual memory cells contain pixel values of information representation. Storage elements of virtual memory cells are classified as either fixed or modifiable so that arbitrary variation of the latter does not cause image destruction. The number of modifiable storage elements juxtaposed with image pixels is regarded as the integer-valued information quantity. A comparison of image information quantity estimations is curried out. The applications of the model to image recognition, document protection, and other tasks are discussed. The text was submitted by the author in English. Mikhail V. Kharinov. Born 1953. Graduated from Leningrad State University in 1978. Received candidateís degree in 1993. Senior researcher at and St. Petersburg Institute of Informatics and Automation, Russian, Academy of Sciences. Scientific interests: analysis of numerical information, system of numerical representation, hierarchical data structures, idempotent transformations, unified algorithms for processing of images and audio signals, color transformation of images. Author of 60 papers, including the patents.     

形状的全尺度可视化表示与识别

目的视觉目标的形状特征表示和识别是图像领域中的重要问题。在实际应用中,视角、形变、遮挡和噪声等干扰因素造成识别精度较低,且大数据场景需要算法具有较高的学习效率。针对这些问题,本文提出一种全尺度可视化形状表示方法。方法在尺度空间的所有尺度上对形状轮廓提取形状的不变量特征,获得形状的全尺度特征。将获得的全部特征紧凑地表示为单幅彩色图像,得到形状特征的可视化表示。将表示形状特征的彩色图像输入双路卷积网络模型,完成形状分类和检索任务。结果通过对原始形状加入旋转、遮挡和噪声等不同干扰的定性实验,验证了本文方法具有旋转和缩放不变性,以及对铰接变换、遮挡和噪声等干扰的鲁棒性。在通用数据集上进行形状分类和形状检索的定量实验,所得准确率在不同数据集上均超过对比算法。在MPEG-7数据集上精度达到99.57%,对比算法的最好结果为98.84%。在铰接和射影变换数据集上皆达到100%的识别精度,而对比算法的最好结果分别为89.75%和95%。结论本文提出的全尺度可视化形状表示方法,通过一幅彩色图像紧凑地表达了全部形状信息。通过卷积模型既学习了轮廓点间的形状特征关系,又学习了不同尺度间的形状特征关系。本文方法...     

复数基下的图像伪装算法

目的 图像的表达始终是图像处理领域的基础问题之一,本文研究了数字图像表达为一个复数的方法, 进一步将这种表示方法应用于图像伪装.方法 在位值制记数法中,取复数为基,将数字图像在复数基下表达为一个高斯整数,从而将图像与复平面上的点建立对应关系.这种数字图像表达方法,能够将图像间的关系转化为复平面上点之间的关系,从而借助几何学方法研究图像问题.据此实现了一种图像伪装算法.结果 该算法不仅可以伪装一幅图像,还可以同时伪装多幅图像,也可以进行一幅图像的多次伪装.结论 实验检测结果表明该伪装算法数值稳定,在理想传输环境下可准确恢复图像,如果传输过程中受到噪声污染,也能较好地恢复保密图像.     

Secure Group Communication with Hidden Group Key

ABSTRACT

Grid environment is mainly concerned with sharing diverse resources belonging to different administrative domains. Security and scalability are important issues for secure group communication in a grid environment. Secure group communication among group members working on collaborative tasks involves confidentiality and authenticity of the message and the session key. In order to transmit the data in a secure way, a suitable key transfer protocol should be implemented. Key transfer protocols rely on a centralized Key Management Center (KMC), an entity responsible for managing user authentication, generating group keys and distributing it to all the members of the communicating group secretly. In this paper, we proposed a secure key transfer protocol that uses the concepts of soft dipole representation of the image and steganography for transferring password and group key securely to the users. In this protocol, users submit their images at the time of registration with the KMC. First KMC compute the soft dipole representation S_I of every user's image and each user's image can be uniquely represented by its corresponding soft dipole representation. The soft dipole representation of each user's image is used as a password for authentication of corresponding user with KMC. KMC generates the group key using soft dipoles of the images of the group members. To preserve the confidentiality, the group key is embedded into an image (using information hiding technique) and KMC broadcast that image to all the group members.     

非负局部Laplacian稀疏编码和上下文信息的图像分类

目的 稀疏编码是图像特征表示的有效方法,但不足之处是编码不稳定,即相似的特征可能会被编码成不同的码字。且在现有的图像分类方法中,图像特征表示和图像分类是相互独立的过程,提取的图像特征并没有有效保留图像特征之间的语义联系。针对这两个问题,提出非负局部Laplacian稀疏编码和上下文信息的图像分类算法。方法 图像特征表示包含两个阶段,第一阶段利用非负局部的Laplacian稀疏编码方法对局部特征进行编码,并通过最大值融合得到原始的图像表示,从而有效改善编码的不稳定性;第二阶段在所有图像特征表示中随机选择部分图像生成基于上下文信息的联合空间,并通过分类器将图像映射到这些空间中,将映射后的特征表示作为最终的图像表示,使得图像特征之间的上下文信息更多地被保留。结果 在4个公共的图像数据集Corel-10、Scene-15、Caltech-101以及Caltech-256上进行仿真实验,并和目前与稀疏编码相关的算法进行实验对比,分类准确率提高了约3%~18%。结论 本文提出的非负局部Laplacian稀疏编码和上下文信息的图像分类算法,改善了编码的不稳定性并保留了特征之间的相互依赖性。实验结果表明,该算法与现有算法相比的分类效果更好。另外,该方法也适用于图像分割、标注以及检索等计算机视觉领域的应用。     

Visual search and preferences concerning different types of guidance displays

This research evaluates the influence of different designs of in car guidance information on visual explorations and subjective preferences, taking into account the drivers characteristics. The results showed that each guidance information representation factors- the number of exit roads, the presence or absence of a landmark and the types of intersections- has a significant influence on the mean maximum duration of glances. Moreover, below a raised threshold of complexity, the number of glances remains stable and independent of the quantity of information presented, whereas passing this threshold demands extra glances from the driver. Lastly, the majority of drivers prefers the complex guidance designs, enabling them to form an unambiguous mental representation. Some effects of the age and gender factors are found. The overall results are interpreted in the framework of the perceptual processes and visuo-spatial working memory models. In conclusion, psycho-ergonomics recommendations are proposed.     

Recognition of dynamic visual images based on group transformations

A new dynamic image recognition method for image sequences based on two-level recognition of individual states and a set of states as a whole is considered in the paper. Integrated and invariant estimations of internal automorphisms are suggested. The modified collective decision-making method for dynamic image recognition uses four types of pseudodistances, in order to obtain a measure of similarity of input dynamic images with dynamic reference images depending on representation of dynamic features, as sets of numerical parameters, sets of vectors, or sets of functions. The experimental data on dynamic image recognition based on various measures of similarity of the sample with the reference pattern, which take into account isomorphic and permissible homomorphic transformations of visual projections, are presented.     

Graph-based markerless registration of city maps using geometric hashing

Recently, augmenting paper maps with additional dynamic information on mobile devices has become popular. A central task in this context is to register high-resolution paper maps to digital maps on a mobile device, which was typically performed by means of RFID tags or visual markers on specially prepared paper maps. In this paper we present a novel graph-based approach for a markerless registration of city maps. The goal is to find the best registration between a given image, which shows a small part of a city map, and stored map data. The proposed method creates a graph representation of a given input image and robustly finds an optimal registration using a geometric hashing technique. It is translation, scale and rotation invariant, and robust against noise and missing data. Experiments on both synthetic and real data are presented to demonstrate the algorithmic performance.     

双重字典学习与自适应PCNN相结合的医学图像融合

目的 针对基于稀疏编码的医学图像融合方法存在的细节保存能力不足的问题,提出了一种基于卷积稀疏表示双重字典学习与自适应脉冲耦合神经网络（PCNN）的多模态医学图像融合方法。方法 首先通过已配准的训练图像去学习卷积稀疏与卷积低秩子字典,在两个字典下使用交替方向乘子法（ADMM）求得其卷积稀疏表示系数与卷积低秩表示系数,通过与对应的字典重构得到卷积稀疏与卷积低秩分量;然后利用改进的的拉普拉斯能量和（NSML）以及空间频率和（NMSF）去激励PCNN分别对卷积稀疏与卷积低秩分量进行融合;最后将融合后的卷积稀疏与卷积低秩分量进行组合得到最终的融合图像。结果 对灰度图像与彩色图像进行实验仿真并与其他融合方法进行比较,实验结果表明,所提出的融合方法在客观评估和视觉质量方面明显优于对比的6种方法,在4种指标上都有最优的表现;与6种多模态图像融合方法相比,3组实验平均标准差分别提高了7%、10%、5.2%;平均互信息分别提高了33.4%、10.9%、11.3%;平均空间频率分别提高了8.2%、9.6%、5.6%;平均边缘评价因子分别提高了16.9%、20.7%、21.6%。结论 与其他稀疏表示方法相比,有效提高了多模态医学图像融合的质量,更好地保留了源图像的细节信息,使融合图像的信息更加丰富,符合人眼的视觉特性,有效地辅助医生进行疾病诊断。     

Quaternion angular radial transform and properties transformation for color-based object recognition

Nowadays, with the increased use of digital images, almost all of which are in color format. Conventional methods process color images by converting them into gray scale, which is definitely not effective in representing and which may lose some significant color information. Recently, a novel method of the Color Angular Radial Transform (CART) is presented. This transform combines the information by considering the shape information inherent in the color. However, ART is adapted on the MPEG-7 standard is only limited to binary images and gray-scale images has many properties: invariant to rotation, Translation and scaling, ability to describe complex objects, so it cannot handle color images directly. To solve this problem we proposed in this article to generalize ART from complex domain to hypercomplex domain using quaternion algebras to achieve the Quaternion Angular Radial Transform (QART) to describe finally two dimensional color images and to insure these properties robustness to all possible rotations and translation and scaling. The performance of QART is then evaluated with large database of color image as an example. We first provide a general formula of ART from which we derive a set of quaternion-valued QART properties transformations by eliminating the influence of transformation parameters. The experimental results show that the QART performs better than the commonly used Quaternion form Zernike Moment (QZM) in terms of image representation capability and numerical stability.     

Multi-image steganography and authentication using crypto-stego techniques

It is a necessity to protect sensitive information in digital form from an adversary who may indulge in cyber-crimes such as modification, masquerading, and replaying of data. Security systems designed to counter such attacks must keep abreast of the adversary. In this paper, we have proposed a novel multi-image crypto-stego technique using Rabin cryptosystem and Arnold transform that provides a mechanism to hide digital data in the form of text, image, audio, and video. The proposed technique is a novel approach for (n,n) secret sharing that prevents attack by an intruder impersonating as a shareholder. In the proposed technique, the header information is created to retrieve data in the correct order. Randomized encrypted data and partial header information are camouflaged in the edges of multiple images in an adaptive manner. Minimal and distribution sequence keys distribute data in shares. Experimental results yield high values of PSNR and low values of MSE for the audio, image, video signals. Further, as the entropy values for original cover image coincide with the crypto-stego image up to the third place of decimal, the secret message will go unnoticed. Sensitivity analysis reveals that even a minor variation in a single share makes the recovery of the secret message infeasible. Comparison with the state of the art techniques indicates that the proposed technique either scores over its competitors or performs equally well in terms of standard evaluation metrics.

     

Selective medical image encryption using DNA cryptography

ABSTRACT

In the medical field, advanced techniques like e-health, smart health, and telemedicine applications are in use. These techniques transmit a digital medical image via open-source networks. The digital medical image contains confidential and sensitive information of patients. The transmitted digital medical images are used for diagnosis in the remote center. Hence, providing security and maintaining the confidentiality of the medical image is a major apprehension. In this paper, DNA cryptography and dual hyperchaotic map techniques are proposed to provide high-level security for a digital medical image. The digital medical images are very large in size and require more computational time. To reduce computational time, the selective digital medical image encryption algorithm is proposed. In the proposed cryptosystem, the permutation and diffusion process are performed on selected pixels of digital medical images. To construct theDNA structure for digital medical images, all DNA encoding rules based on the pixel position of the digital medical image are used. The cipher image is attained by using all DNA decoding rules based on the pixel value of the digital medical image. The proposed cryptosystem is resistant to different types of attacks.     

面向可逆图像处理网络的可证安全自然隐写

目的 自然隐写是一种基于载体源转换的图像隐写方法,基本思想是使隐写后的图像具有另一种载体的特征,从而增强隐写安全性。但现有的自然隐写方法局限于对图像ISO(International Standardization Organization)感光度进行载体源转换,不仅复杂度高,而且无法达到可证安全性。为了提高安全性,本文结合基于标准化流的可逆图像处理模型,在隐空间完成载体源转换,同时通过消息映射的设计做到了可证安全的自然隐写。方法 利用目前发展迅速的基于可逆网络的图像处理方法将图像可逆地映射到隐空间,通过替换使用的隐变量完成载体源的转换,从而避免对原始图像复杂的建模。同时,改进了基于拒绝采样的消息映射方法,简单地从均匀分布中采样以获得需要的条件分布,高效地将消息嵌入到隐变量中,并且保证了嵌入消息后的分布与原本使用的分布一致,从而实现了可证安全的自然隐写。结果 针对图像质量、隐写容量、消息提取准确率、隐写安全性和运行时间进行了实验验证,结果表明在使用可逆缩放网络和可逆去噪网络时能够在每个像素值上平均嵌入5.625 bit消息,且具有接近99%的提取准确率,同时隐写分析网络SRNet(st...     

自学习规则下的多聚焦图像融合

目的 基于深度学习的多聚焦图像融合方法主要是利用卷积神经网络（convolutional neural network,CNN）将像素分类为聚焦与散焦。监督学习过程常使用人造数据集,标签数据的精确度直接影响了分类精确度,从而影响后续手工设计融合规则的准确度与全聚焦图像的融合效果。为了使融合网络可以自适应地调整融合规则,提出了一种基于自学习融合规则的多聚焦图像融合算法。方法 采用自编码网络架构,提取特征,同时学习融合规则和重构规则,以实现无监督的端到端融合网络;将多聚焦图像的初始决策图作为先验输入,学习图像丰富的细节信息;在损失函数中加入局部策略,包含结构相似度（structural similarity index measure,SSIM）和均方误差（mean squared error,MSE）,以确保更加准确地还原图像。结果 在Lytro等公开数据集上从主观和客观角度对本文模型进行评价,以验证融合算法设计的合理性。从主观评价来看,模型不仅可以较好地融合聚焦区域,有效避免融合图像中出现伪影,而且能够保留足够的细节信息,视觉效果自然清晰;从客观评价来看,通过将模型融合的图像与其他主流多聚焦图像融合算法的融合图像进行量化比较,在熵、Q_w、相关系数和视觉信息保真度上的平均精度均为最优,分别为7.457 4,0.917 7,0.978 8和0.890 8。结论 提出了一种用于多聚焦图像的融合算法,不仅能够对融合规则进行自学习、调整,并且融合图像效果可与现有方法媲美,有助于进一步理解基于深度学习的多聚焦图像融合机制。     

一种基于卷积神经网络的砂岩显微图像特征表示方法

砂岩显微图像分类是地质学研究中一项基本工作,在油气储集层评估等方面有重要意义.在实现自动分类时,由于砂岩显微图像具有复杂多变的显微结构,人工定义特征对砂岩显微图像的表示能力有限.此外,由于样本采集和标注成本高昂,带标记的砂岩显微图像很少.提出一种面向小规模数据集的基于卷积神经网络的特征表示方法FeRNet,以便有效地捕获砂岩显微图像的语义信息,提高对砂岩显微图像的特征表示能力.FeRNet网络结构简单,可降低网络对带标记图像数据量的要求,防止参数过拟合.针对带标记砂岩显微图像数量不足的问题,提出了图像扩增预处理方法及基于卷积自编码网络的权重初始化策略,降低了因数据不足造成的过拟合风险.基于采自西藏地区的砂岩显微图像数据集设计并进行实验,实验结果表明,在带标记砂岩显微图像数据不足的情况下,图像扩增和卷积自编码网络可以有效地改善FeRNet网络的训练效果,通过FeRNet网络提取的特征对砂岩显微图像的表示能力优于人工定义特征.     

卷积稀疏与细节显著图解析的图像融合

目的 针对图像融合中信息量不够丰富和边缘细节模糊问题,结合多尺度分析、稀疏表示和显著性特征等图像表示方法,提出一种卷积稀疏与细节显著图解析的图像融合方法。方法 首先构造一种自适应样本集,训练出契合度更高的字典滤波器组。然后将待融合图像进行多尺度分析得到高低频子图,对低频子图进行卷积稀疏表示,通过权重分析构建一种加权融合规则,得到信息量更加丰富的低频子图;对高频子图构造细节显著图,进行相似性分析,建立一种高频融合规则,得到边缘细节更加凸显的高频子图。最后进行相应逆变换得到最终图像。结果 实验在随机挑选的3组灰度图像集和4组彩色图像集上进行,与具有代表性的7种融合方法进行效果对比。结果表明,本文方法的视觉效果明显较优,平均梯度上依次平均提高39.3%、32.1%、34.7%、28.3%、35.8%、28%、30.4%;在信息熵上依次平均提高6.2%、4.5%、1.9%、0.4%、1.5%、2.4%、2.9%;在空间频率上依次平均提高31.8%、25.8%、29.7%、22.2%、28.6%、22.9%、25.3%;在边缘强度上依次平均提高39.5%、32.1%、35.1%、28.8%、36.6%、28.7%、31.3%。结论 本文方法在一定程度上解决了信息量不足的问题,较好地解决了图像边缘细节模糊的问题,使图像中奇异性更加明显的内容被保留下来。     

基于双重L2稀疏编码的高光谱图像分类

目的 为了有效提高高光谱图像分类的精度,提出了双重L2稀疏编码的高光谱图像分类方法。方法 首先对高光谱图像进行预处理,充分结合图像的空间信息和光谱信息,利用像元的空间连续性,用L2稀疏编码重建图像中每个像元。针对重建的图像数据,依据L2稀疏编码的最小误差和编码系数实现分类。结果 在公开的数据库AVIRIS高光谱图像上进行验证,分类精度为99.44%,与支持向量机（SVM）、K最近邻（KNN）和L1稀疏编码方法比较,有效地提高了分类的准确性。结论 实验结果表明,提出的方法应用于高光谱图像分类具有较好的分类效果。     

A wavelet based multiresolution algorithm for rotation invariant feature extraction

The present work aims at proposing a new wavelet representation formula for rotation invariant feature extraction. The algorithm is a multilevel representation formula involving no wavelet decomposition in standard sense. Using the radial symmetry property, that comes inherently in the new representation formula, we generate the feature vectors that are shown to be rotation invariant. We show that, using a hybrid data mining technique, the algorithm can be used for rotation invariant content based image retrieval (CBIR). The proposed rotation invariant retrieval algorithm, suitable for both texture and nontexture images, avoids missing any relevant images but may retrieve some other images which are not very relevant. We show that the higher precision can however be achieved by pruning out irrelevant images.     

无损载体和鲁棒代价结合的JPEG图像鲁棒隐写

目的传统隐写技术在实际社交网络信道上难以保护秘密信息的完整性。在社交网络中,图像往往经过有损压缩信道进行传输,从而导致隐蔽通信失效。为了保证经过压缩信道传输的载密图像鲁棒性,设计安全鲁棒的隐蔽通信技术具有实际应用价值。基于最小化图像信息损失,本文提出无损载体和鲁棒代价结合的JPEG图像鲁棒隐写。方法首先,指出构造无损载体能有效维持隐写安全性和鲁棒性的平衡,对经过压缩信道前后的JPEG图像空域像素块进行差分,构造无损载体以确定鲁棒嵌入域;其次,通过对离散余弦变换(discrete cosine transform, DCT)系数进行"±1"操作,并计算空域信息在压缩传输前后的损失,设计衡量DCT系数抗压缩性能的鲁棒代价;同时,验证在低质量因子压缩信道下鲁棒代价更能区分DCT系数的鲁棒能力,最后,利用校验子格编码(syndrome-trellis code, STC),结合无损载体和鲁棒代价对秘密信息进行嵌入。结果实验在BossBase1.01图像库上进行对比实验,相比于传统JPEG隐写技术,构造无损载体作为嵌入域能有效地将信息平均提取错误率降低24.97%,图像的正确提取成功率提高了21...