Image segmentation and image models

This paper discusses image segmentation techniques from the standpoint of the assumptions that an image should satisfy in order for a particular technique to be applicable to it. These assumptions, which are often not stated explicitly, can be regarded as (perhaps informal) "models" for classes of images. The paper emphasizes two basic classes of models: statistical models that describe the pixel population in an image or region, and spatial models that describe the decomposition of an image into regions.     

Hierarchical Feature Extraction for Image Recognition

This paper applies a hierarchical classifier to two image recognition tasks. At the heart of this classifier, like many other classifiers, is a distance metric for determining the similarity of pairs of images. As the generalisation performance is often strongly related to the effectiveness of this measure, this paper develops a measure that is statistically more reliable than some metrics, but does not discard discriminating information, often regarded as noise. In addition, it may be computed quickly. This paper also experimentally shows that the metric may be used in the hierarchical classifier to yield error rates far lower to those based on the Euclidean distance metric on the two image recognition tasks. Furthermore, it gives the lowest reported error rate (2.63%) as well as the best training and classification times for a face recognition task.     

Automatic background recognition and removal (ABRR) in computedradiography images

A novel method to automatically recognize and remove background signals in computed radiography (CR) images caused by X-ray collimation during projection radiographic examinations is presented. There are three major steps in this method. In the first step, a statistical curve is derived based on many hierarchical CR sample images as a first approximation to loosely separate image and background pixels. Second, signal processing methods, including specific sampling, filtering, and angle recognition, are used to determine edges between image and background pixels. Third, adaptive parameter adjustments and consistent and reliable estimation rules are used to finalize the location of edges and remove the background. In addition, this step also evaluates the reliability of the complete background removal operation. With this novel method implemented in a clinical picture archiving and communication system (PACS) at the University of California at San Francisco, the authors achieved 99% correct recognition of CR image background, and 91% full background removal without removing any valid image information     

基于不完全分层MRF的非监督图象分割

分层马尔可夫随机场(MRF)图象模型由于层间具有因果关系,且这种因果关系符合图象的性质,使基于该模型的图象处理时间比平面MRF模型所用的时间大为减少.针对作者提出的一种新的分层马尔可夫图象模型——不完全分层模型,导出EM算法以估计模型参数.算法继承了分层模型非迭代算法运算速度快的优点,并因为模型结构的简化进一步减少了计算量,算法在模型的最上层加入了平面节点间信息的交互,以较少的计算换来了更加精确的参数估计结果.算法用于图象非监督分割的实验表明,和分层模型算法相比,其处理速度更快、由所估计的参数得到了更好或相当的分割结果,尤其适合大幅面图象的处理.     

A Markov model for blind image separation by a mean-field EM algorithm.

This paper deals with blind separation of images from noisy linear mixtures with unknown coefficients, formulated as a Bayesian estimation problem. This is a flexible framework, where any kind of prior knowledge about the source images and the mixing matrix can be accounted for. In particular, we describe local correlation within the individual images through the use of Markov random field (MRF) image models. These are naturally suited to express the joint pdf of the sources in a factorized form, so that the statistical independence requirements of most independent component analysis approaches to blind source separation are retained. Our model also includes edge variables to preserve intensity discontinuities. MRF models have been proved to be very efficient in many visual reconstruction problems, such as blind image restoration, and allow separation and edge detection to be performed simultaneously. We propose an expectation-maximization algorithm with the mean field approximation to derive a procedure for estimating the mixing matrix, the sources, and their edge maps. We tested this procedure on both synthetic and real images, in the fully blind case (i.e., no prior information on mixing is exploited) and found that a source model accounting for local autocorrelation is able to increase robustness against noise, even space variant. Furthermore, when the model closely fits the source characteristics, independence is no longer a strict requirement, and cross-correlated sources can be separated, as well.     

Reversible data-hiding for progressive image transmission

This research presents a multi-resolution reversible data-hiding algorithm to enable multi-scale marked images that are transmitted progressively to be exactly recovered at the receiver side once hidden data has been extracted. Based on the spatially hierarchical multi-layer structures of progressive-image transmission, the proposed algorithm first decimates the incoming image pixels into a pre-specified number of hierarchical layers of pixels. Then, it modifies pixel values in each hierarchical layer by shifting the interpolated-difference-values histogram between two neighboring layers of pixels to embed secret information into the corresponding hierarchical layer images. The proposed algorithm offers a reversible data-hiding ability for applications that use progressive image transmission to render progressive-image authentication, information-tagging, covert communications, etc. With progressive-reversible data-hiding, users of progressive image transmission can receive each original progressive image and complete hidden messages related to the received progressive image. This allows users to make real-time definite decisions according to an application's requirements. In contrast to other reversible data-hiding schemes, the algorithm proposed in this study features reversible data-hiding in progressive-image transmission based on a hierarchical decimation and interpolation technique. The interpolating process is used to reduce the difference values between the target pixel values in one progressive layer and their interpolated ones. This increases the hiding capacity of interpolation-differences histogram shifting. The experimental results demonstrate that the proposed method provides a greater embedding capacity and maintains marked images at a higher quality. Moreover, the proposed method has a low computational complexity as it requires only simple arithmetic computations.     

Preserving step edges in low bit rate progressive image compression

With the growing importance of low-bandwidth applications, such as wireless access to the Internet, images are often sent or received at low bit rates. At these bit rates, they suffer from significant distortion and artifacts, making it difficult for those viewing the images to understand them. We present two progressive compression algorithms that focus on preserving the clarity of important image features, such as edges, at compression ratios of 80:1 and more. Both algorithms capture and encode the locations of important edges in the images. The first algorithm then transmits a standard SPIHT (set partitioning in hierarchical trees) bit stream, and at the decoder applies a nonlinear edge-enhancement procedure to improve the clarity of the encoded edges. The second approach uses a modified wavelet transform to "remove" the edges, and encodes the remaining texture information using SPIHT. With both approaches, features in the images that may be important for recognition are well preserved, even at low bit rates.     

基于二维压缩感知和分层特征的图像检索算法

为了保留图像分析时的像素点位置关系及降维处理,把一维压缩感知理论推广到二维,建立了二维可稀疏信号的压缩测量模型,研究了一种二维信号的自适应梯度下降重构AGDR(Adaptive Gradient Descent Recursion)算法,由此提出了一种图像分层特征提取与检索方法.首先对图像在RGB颜色空间上进行网格离散划分,通过分层算子对图像进行分层映射,定义一种基于颜色网格空间的扩展灰度共生矩阵,采用二维测量模型获取图像的分层测量特征、纹理特征与分层颜色统计特征,图像分层测量特征综合反映出图像的颜色及像素点位置的关系,扩展灰度共生矩阵反映纹理特征.其次用AGDR算法计算检索图像之间的原始信号差量及其稀疏值.最后结合两类分层特征差量、稀疏值和颜色统计特征,融合计算图像间整体相似度度量指标.仿真实验表明,应用分层二维压缩感知测量与AGDR算法的图像检索方法在检索时间、查全率和查准率等指标上具有优越性能,为图像检索提供了新思路.     

Unsupervised statistical neural networks for model-based objectrecognition

Statistical neural networks executing soft-decision algorithms have been shown to be very effective in many classification problems. A neural network architecture is developed here that can perform unsupervised joint segmentation and labeling of objects in images. We propose the semi-parametric hierarchical mixture density (HMD) model as a tool for capturing the diversity of real world images and pose the object recognition problem as a maximum likelihood (ML) estimation of the HMD parameters. We apply the expectation-maximization (EM) algorithm for this purpose and utilize ideas and techniques from statistical physics to cast the problem as the minimization of a free energy function. We then proceed to regularize the solution thus obtained by adding smoothing terms to the objective function. The resulting recursive scheme for estimating the posterior probabilities of an object's presence in an image corresponds to an unsupervised feedback neural network architecture. We present here the results of experiments involving recognition of traffic signs in natural scenes using this technique     

Accelerating rotation of high-resolution images

Real-time image rotation is an essential operation in many application areas such as image processing, computer graphics and pattern recognition. Existing architectures that rely on CORDIC computations for trigonometric operations cause a severe bottleneck in high-throughput applications, especially where high-resolution images are involved. A novel hierarchical method that exploits the symmetrical characteristics of the image to accelerate the rotation of high-resolution images is presented. Investigations based on a 512times512 image show that the proposed method yields a speedup of ~20times for a mere 3% increase in area cost when compared with existing techniques. Moreover, the effect of hierarchy on the computational efficiency has been evaluated to provide for area-time flexibility. The proposed technique is highly scalable and significant performance gains are evident for very high-resolution images     

面向统计特性一致的深度图像修复方法

本文提出了一种改善深度修复图像统计特性一致性的方法。首先,分别采用非线性高通滤波残差及深度神经网络提取固有身份信号(intrinsic identity signal, IIS),发现深度修复图像和真实图像存在IIS统计特性差异,验证在不同来源图像和不同的深度修复算法的条件下统计特性不一致性是普遍存在的。其次,提出一个生成型卷积神经网络,优化修复区域,保证修复图像的视觉质量,使其与真实区域IIS统计特性保持一致。最后,通过在合理范围内对生成网络的部分参数进行随机扰动,生成具有模式多样性的图像,有效降低生成图像被识别来源的概率。通过对比真实图像、深度修复图像、生成图像的IIS统计特性,以及在取证检测器上的对抗检测实验,表明了本文方法的有效性。      

Design and statistical analysis of a hash-aided image watermarking system

This paper develops a joint hashing/watermarking scheme in which a short hash of the host signal is available to a detector. Potential applications include content tracking on public networks and forensic identification. The host data into which the watermark is embedded are selected from a secret subset of the full-frame discrete cosine transform of an image, and the watermark is inserted through multiplicative embedding. The hash is a binary version of selected original image coefficients. We propose a maximum likelihood watermark detector based on a statistical image model. The availability of a hash as side information to the detector modifies the posterior distribution of the marked coefficients. We derive Chernoff bounds on the receiver operating characteristic performance of the detector. We show that host-signal interference can be rejected if the hash function is suitably designed. The relative difficulty of an eavesdropper's detection problem is also determined; the eavesdropper does not know the secret key used. Monte Carlo simulations are performed using photographic test images. Finally, various attacks on the watermarked image are introduced to study the robustness of the derived detectors. The joint hashing/watermarking scheme outperforms the traditional "hashless" watermarking technique.     

Recognition of dynamic video contents with global probabilistic models of visual motion.

The exploitation of video data requires methods able to extract high-level information from the images. Video summarization, video retrieval, or video surveillance are examples of applications. In this paper, we tackle the challenging problem of recognizing dynamic video contents from low-level motion features. We adopt a statistical approach involving modeling, (supervised) learning, and classification issues. Because of the diversity of video content (even for a given class of events), we have to design appropriate models of visual motion and learn them from videos. We have defined original parsimonious global probabilistic motion models, both for the dominant image motion (assumed to be due to the camera motion) and the residual image motion (related to scene motion). Motion measurements include affine motion models to capture the camera motion and low-level local motion features to account for scene motion. Motion learning and recognition are solved using maximum likelihood criteria. To validate the interest of the proposed motion modeling and recognition framework, we report dynamic content recognition results on sports videos.     

Gabor-based kernel PCA with doubly nonlinear mapping for face recognition with a single face image.

In this paper, a novel Gabor-based kernel principal component analysis (PCA) with doubly nonlinear mapping is proposed for human face recognition. In our approach, the Gabor wavelets are used to extract facial features, then a doubly nonlinear mapping kernel PCA (DKPCA) is proposed to perform feature transformation and face recognition. The conventional kernel PCA nonlinearly maps an input image into a high-dimensional feature space in order to make the mapped features linearly separable. However, this method does not consider the structural characteristics of the face images, and it is difficult to determine which nonlinear mapping is more effective for face recognition. In this paper, a new method of nonlinear mapping, which is performed in the original feature space, is defined. The proposed nonlinear mapping not only considers the statistical property of the input features, but also adopts an eigenmask to emphasize those important facial feature points. Therefore, after this mapping, the transformed features have a higher discriminating power, and the relative importance of the features adapts to the spatial importance of the face images. This new nonlinear mapping is combined with the conventional kernel PCA to be called "doubly" nonlinear mapping kernel PCA. The proposed algorithm is evaluated based on the Yale database, the AR database, the ORL database and the YaleB database by using different face recognition methods such as PCA, Gabor wavelets plus PCA, and Gabor wavelets plus kernel PCA with fractional power polynomial models. Experiments show that consistent and promising results are obtained.     

分层级联生成对抗网络用于手背静脉图像修复

为解决在识别过程中因手背静脉图像信息缺失而造成识别效率低下的问题,本文提出了分层级联生成对抗网络的手背静脉图像修复框架。该网络框架分别以级联与并行分层的方式进行修复操作,通过并行分层结构创新性的融合了不同静脉图像的特征信息;为有效地利用静脉图像的上下文信息对缺失的静脉图像信息进行预测与补全,在网络中创新性的引入了空洞卷积核与非局部注意力网络;为保证修复静脉图像质量与其真实图像的一致性,创新性的结合对抗损失与感知损失进行优化。实验结果表明,本文算法在视觉效果、峰值信噪比（Peak Signal to Noise Ratio,PSNR）和结构相似性（Structural Similarity Index,SSIM）等方面表现优于已有算法,并在两个公开的掌纹与指纹数据集上进行了有效的泛化验证。此外,修复图像相较于缺失图像在身份识别效率方面有了一定的提高。      

一种基于帧间差分法的舰船中靶图像识别方法

将帧间差分法运用于舰船中靶图像识别中,并提出了一种基于图像统计特征的舰船中靶图像识别方法。基于Visual C＋＋中的MFC对话框开发工具,借助OpenCV计算机视觉库编写了演示软件,并对识别方法进行了验证。结果表明,该方法能快速、准确地识别出中靶图像,从而为指挥员的指挥决策提供客观依据。     

A framework for evaluating the data-hiding capacity of image sources

An information-theoretic model for image watermarking and data hiding is presented in this paper. Previous theoretical results are used to characterize the fundamental capacity limits of image watermarking and data-hiding systems. Capacity is determined by the statistical model used for the host image, by the distortion constraints on the data hider and the attacker, and by the information available to the data hider, to the attacker, and to the decoder. We consider autoregressive, block-DCT, and wavelet statistical models for images and compute data-hiding capacity for compressed and uncompressed host-image sources. Closed-form expressions are obtained under sparse-model approximations. Models for geometric attacks and distortion measures that are invariant to such attacks are considered.     

Including efficient object recognition capabilities in online robots: from a statistical to a Neural-network classifier

For those situations in which the user wants to interact with the system by using, for example, voice commands, it would be convenient to refer to the objects by their names (e.g., "cube") instead of other types of interactions (e.g., "grasp object 1"). Thus, automatic object recognition is the first step in order to acquire a higher level of interaction between the user and the robot. Nevertheless, applying object recognition techniques when the camera images are being transmitted through the web is not an easy task. In this situation, images cannot have a very high resolution, which affects enormously the recognition process due to the inclusion of more errors while digitalizing the real image. Some experiments with the Universitat Jaume I Online Robot evaluate the performance of different neural-network implementations, comparing it to that of some distance-based object recognition algorithms. Results will show which combination of object features, and algorithms (both statistical and neural networks) is more appropriate to our purpose in terms of both effectiveness and computing time.