Non-Rigid Image Registration by Neural Computation

Non-rigid image registration is a prerequisite for many medical image analysis applications such as image fusion of multimodality images and quantitative change analysis of a temporal sequence in computer-aided diagnosis. By establishing the point correspondence of the extracted feature points, it is possible to recover the deformation using nonlinear interpolation methods. However, it may be very difficult to establish such correspondence at an initial stage when confronted with large and complex deformation. In this paper, a mixture of principal axes registration (mPAR) method is proposed to resolve the correspondence problem through a neural computational approach. The novel feature of mPAR is the alignment of two point sets without the need of establishing explicit point correspondence. Instead, it aligns the two point sets by minimizing the relative entropy between their probability distributions resulting in a maximum likelihood estimate of the transformation matrix. The registration process consists of two steps: (1) a finite mixture scheme to establish an improved point correspondence and (2) a multilayer perceptron neural network (MLP) to recover the nonlinear deformation. The neural computation for registration uses a committee machine to obtain a mixture of piece-wise rigid registrations, which gives a reliable point correspondence using multiple extracted objects in a finite mixture scheme. Then the MLP is used to determine the coefficients of a polynomial transform using extracted control points. We have applied our mPAR method to register synthetic data sets, surgical prostate models, and a temporal sequence of mammograms of a single patient. The experimental results show that mPAR not only improves the accuracy of the point correspondence but also results in a desirable error-resilience property for control point selection errors.     

Use of multiresolution wavelet feature pyramids for automatic registration of multisensor imagery.

The problem of image registration, or the alignment of two or more images representing the same scene or object, has to be addressed in various disciplines that employ digital imaging. In the area of remote sensing, just like in medical imaging or computer vision, it is necessary to design robust, fast, and widely applicable algorithms that would allow automatic registration of images generated by various imaging platforms at the same or different times and that would provide subpixel accuracy. One of the main issues that needs to be addressed when developing a registration algorithm is what type of information should be extracted from the images being registered, to be used in the search for the geometric transformation that best aligns them. The main objective of this paper is to evaluate several wavelet pyramids that may be used both for invariant feature extraction and for representing images at multiple spatial resolutions to accelerate registration. We find that the bandpass wavelets obtained from the steerable pyramid due to Simoncelli performs best in terms of accuracy and consistency, while the low-pass wavelets obtained from the same pyramid give the best results in terms of the radius of convergence. Based on these findings, we propose a modification of a gradient-based registration algorithm that has recently been developed for medical data. We test the modified algorithm on several sets of real and synthetic satellite imagery.     

A rapid and automatic image registration algorithm with subpixelaccuracy

A number of digital imaging techniques in medicine require the combination of multiple images. Using these techniques, it is essential that the images be adequately aligned and registered prior to addition, subtraction, or any other combination of the images. This paper describes an alignment routine developed to register an image of a fixed object containing a global offset error, rotation error, and magnification error relative to a second image. The described routine uses sparsely sampled regional correlation in a novel way to reduce computation time and avoid the use of markers and human interaction. The result is a fast, robust, and automatic alignment algorithm, with accuracy better than about 0.2 pixel in a test with clinical computed radiography images     

一种基于稠密SIFT特征对齐的稀疏表达人脸识别算法

该文针对人脸图像受到非刚性变化的影响,如旋转、姿态以及表情变化等,提出一种基于稠密尺度不变特征转换(SIFT)特征对齐(Dense SIFT Feature Alignment, DSFA)的稀疏表达人脸识别算法。整个算法包含两个步骤:首先利用DSFA方法对齐训练和测试样本;然后设计一种改进的稀疏表达模型进行人脸识别。为加快DSFA步骤的执行速度,还设计了一种由粗到精的层次化对齐机制。实验结果表明:在ORL,AR和LFW 3个典型数据集上,该文方法都获得了最高的识别精度。该文方法比传统稀疏表达方法在识别精度上平均提高了4.3%,同时提高了大约6倍的识别效率。     

基于近似距离序的重复模式图像匹配算法

由于重复模式图像局部信息的歧义性,即使在视角变化很小的情况下,仅通过比较局部描述子之间的相似性得到的匹配结果往往并不可靠。为了解决这个问题,该文根据特征点空间的分布特性,给出了一种新颖的几何相容性表示方法近似距离序;结合利用局部描述子获得的匹配代价,定义了一种混合形式的目标函数,将匹配问题转化为一对一约束下的优化问题;最后,通过概率松弛法求解目标函数的极大值,获得特征点之间的对应关系。在不同类型图像上的比较实验表明,所提出的方法能够有效地解决重复模式图像匹配问题。     

Robust and accurate registration of 2-D electrophoresis gels using point-matching.

Point-matching is a widely applied image registration method and many algorithms have been developed. Registration of 2-D electrophoresis gels is an important problem in biological research that presents many of the technical difficulties that beset point-matching: large numbers of points with variable densities, large nonrigid transformations between point sets, paucity of structural information and large numbers of unmatchable points (outliers) in either set. In seeking the most suitable algorithm for gel registration we have evaluated a number of approaches for accuracy and robustness in the face of these difficulties. Using synthetic images we test combinations of three algorithm components: correspondence assignment, distance metrics and image transformation. We show that a version of the iterated closest point (ICP) algorithm using a non-Euclidean distance metric and a robust estimation of transform parameters provides best performance, equalling SoftAssign in the presence of moderate image distortion, and providing superior robustness against large distortions and high outlier proportions. From this evaluation we develop a gel registration algorithm based on robust ICP and a novel distance metric combining Euclidean, shape context and image-related features. We demonstrate the accuracy of gel matching using synthetic distortions of real gels and show that robust estimation of transform parameters using M-estimators can enforce inverse consistency, ensuring that matching results are independent of the order of the images.     

Hierarchical browsing and search of large image databases

The advent of large image databases (>10000) has created a need for tools which can search and organize images automatically by their content. This paper focuses on the use of hierarchical tree-structures to both speed-up search-by-query and organize databases for effective browsing. The first part of this paper develops a fast search algorithm based on best-first branch and bound search. This algorithm is designed so that speed and accuracy may be continuously traded-off through the selection of a parameter lambda. We find that the algorithm is most effective when used to perform an approximate search, where it can typically reduce computation by a factor of 20-40 for accuracies ranging from 80% to 90%. We then present a method for designing a hierarchical browsing environment which we call a similarity pyramid. The similarity pyramid groups similar images together while allowing users to view the database at varying levels of resolution. We show that the similarity pyramid is best constructed using agglomerative (bottom up) clustering methods, and present a fast sparse clustering method which dramatically reduces both memory and computation over conventional methods.     

基于边缘特征的SAR图像与光学图像的匹配

提出了一种新的基于边缘特征的合成孔径雷达(SAR)图像与光学图像相匹配的算法，可用于无人飞行器的导航。首先针对SAR图像低信噪比的特定情况，提出了一种简单实用的线状特征提取方法。而后在提取边缘的基础上，采用改进的具有强抗干扰能力和容错能力的Hausdorff距离作为相似性准则进行图像匹配。在搜索策略上．采用遗传算法来加快搜索的速度。实验结果表明，这种算法匹配精度高，计算速度快。     

Robust Visual Tracking with SIFT Features and Fragments Based on Particle Swarm Optimization

We propose a novel approach for visual tracking based on a particle swarm optimization (PSO) framework using SIFT feature points correspondence and multiple fragments in a candidate target region to cope with the problems of partial occlusions, illumination changes, and large motion changes of the tracked target. Firstly, optimal search in the successive frame tracking process is performed by the PSO algorithm, which guides all particles towards the global optima state based on a fitness function. Then, the SIFT feature information is integrated into the iterative results of PSO to acquire a more accurate tracking state. Secondly, we present an effective appearance model updating criterion, which evaluates which fragments in appearance model need updating at each frame. However, the fragments with occluded parts or low quality measure values are not updated. The method for updating appearance model is introduced to improve the tracking performance. Compared with state-of-the-art algorithms, the proposed method can still stably track the target during the course of long-term partial occlusions using superior fragments of tracked target. The experiment results demonstrate the effectiveness of our algorithm in complex environments where the target object undergoes partial occlusions and large changes in pose and illumination.     

Successive elimination algorithm for motion estimation

The correspondence presents a fast exhaustive search algorithm for motion estimation. The basic idea is to obtain the best estimate of the motion vectors by successively eliminating the search positions in the search window and thus decreasing the number of matching evaluations that require very intensive computations. Simulation results demonstrate that although the performance of the proposed algorithm is the same as that using the exhaustive search, the computation time has been reduced significantly     

Correspondence Propagation with Weak Priors

For the problem of image registration, the top few reliable correspondences are often relatively easy to obtain, while the overall matching accuracy may fall drastically as the desired correspondence number increases. In this paper, we present an efficient feature matching algorithm to employ sparse reliable correspondence priors for piloting the feature matching process. First, the feature geometric relationship within individual image is encoded as a spatial graph, and the pairwise feature similarity is expressed as a bipartite similarity graph between two feature sets; then the geometric neighborhood of the pairwise assignment is represented by a categorical product graph, along which the reliable correspondences are propagated; and finally a closed-form solution for feature matching is deduced by ensuring the feature geometric coherency as well as pairwise feature agreements. Furthermore, our algorithm is naturally applicable for incorporating manual correspondence priors for semi-supervised feature matching. Extensive experiments on both toy examples and real-world applications demonstrate the superiority of our algorithm over the state-of-the-art feature matching techniques.      

一种基于参数重估计和分层匹配的PTZ相机参数修正算法

由于多种因素的影响,PTZ(pan-ti lt-zoom)相机经过长时间运行后其参数会偏离真实值,因 此有必要对相机参数进行修正。针对现有算法参数修正精度低、适应能力弱等问题,提 出了一 种改进的基于参数重估计和分层匹配的参数修正算法。算法通过引入参数重估计策略,避免 了参数 修正过程中的误差累积,提高了参数修正的精度;通过设计分层匹配和特征传播步骤,增强 了算法 对不同尺度图像的适应性。实际场景中的多组实验结果表明,本文算法可以准确修正PTZ相 机参数,并且较现有方法更具优越性。     

利用SVM的极化SAR图像特征选择与分类

该文提出一种新的利用SVM的特征选择算法,并将其融入到极化SAR图像分类过程中,构成一种新的基于SVM的分类方法。其中,特征选择算法利用支持向量个数作为特征评估指标,并以顺序后退法作为搜索策略。真实数据的实验结果表明,该分类方法能有效降低SVM分类器对自身参数的敏感性,与利用原始特征集和经典的RELIEF-F的分类方法相比,该方法能以更少(或相当)的特征个数,在更广泛的SVM参数取值范围内获得更高的分类精度。     

基于仿射迭代方法的大基线图像匹配技术研究

两幅图像中相应特征点邻域窗口之间的单应映射可以用仿射变换模型来近似。本文首先通过奇异值分解给出仿射变换矩阵4个自由度的几何含义,然后将其分解为一个相似变换矩阵和一个旋转的准单位矩阵(Rotated Quasi-Identity Matrix)的乘积,即在基于相似变换模型匹配的基础上再用基于仿射变换模型的迭代算法对相应特征点精确定位。针对相似变换中初始旋转角度的难确定性,在初始匹配中提出基于亮度最速下降方向的对齐方法,而在引导匹配阶段提出基于相应极线方向的对齐方法,这两个策略不仅提高了算法效率,还能为进一步的仿射迭代提供良好的初值。在得到最优仿射变换参数之后,实现了对相应特征点定位误差的精确补偿及其邻域窗口的透视矫正。最后通过真实图像的实验以及和现有算法的比较验证了本文算法的可行性和精确性,并给出了相应的实验数据和结果。     

基于SURF特征的栈式自编码网络人脸对齐算法

人脸对齐是人脸识别系统中的一个核心部分,定位的准确性和定位速度直接影响到人脸识别的效果。人脸图像存在不同姿态、不同表情、不同光照条件等因素的影响,真实场景下的人脸对齐成为一个难题。本文提出了一种基于SURF特征的栈式自编码网络人脸对齐方法,首先通过粗糙定位网络找到近似人脸特征点,并提取局部的SURF特征,输入到局部细化网络,通过级联结构,进一步精确化人脸特征点的具体位置。最后,在人脸数据集AFLW和HELEN上与近几年的对齐方法进行对比实验,平均错误率8.80%,i5四核CPU,2.3Hz主频硬件平台下计算时间7.6ms。我们的人脸对齐方法在真实场景下(包括单人和多人)具有较好的鲁棒性,可以实现准确定位。     

Image mosaic with relaxed motion

We propose a novel method to stitch images with relatively large roll or pitch called relaxed motion, which defies most existing mosaic algorithms. Our approach adopts a multi-resolution strategy, which combines the merits of both feature-based and intensity-based methods. The main contribution is a robust motion estimation procedure which integrates an adaptive multi-scale block matching algorithm called TV-BMA, a low contrast filter and a RANSAC motion rectification to jointly refine motion and feature matches. Based on TV ? L ¹ model, the proposed TV-BMA works on the coarsest layer to find a robust initial displacement field as the initial motion for source images. This motion estimation method can generate robust correspondences for further processing. In the subsequent camera calibration step, we also present two stable methods to estimate the camera matrix. To estimate the focal length, we combine the golden section search and the simplex method based on the angle invariance of feature vectors; to estimate the rotation matrix, we introduce a subspace trust region method, which matches features based on the rotation invariance. Extensive experiments show that our approach leads to improved accuracy and robustness for stitching images with relaxed motion.     

An Optical Flow-Based Approach to Robust Face Recognition Under Expression Variations

Face recognition is one of the most intensively studied topics in computer vision and pattern recognition, but few are focused on how to robustly recognize faces with expressions under the restriction of one single training sample per class. A constrained optical flow algorithm, which combines the advantages of the unambiguous correspondence of feature point labeling and the flexible representation of optical flow computation, has been developed for face recognition from expressional face images. In this paper, we propose an integrated face recognition system that is robust against facial expressions by combining information from the computed intraperson optical flow and the synthesized face image in a probabilistic framework. Our experimental results show that the proposed system improves the accuracy of face recognition from expressional face images.     

Image Registration Using Adaptive Polar Transform

Image registration is an essential step in many image processing applications that need visual information from multiple images for comparison, integration, or analysis. Recently, researchers have introduced image registration techniques using the log-polar transform (LPT) for its rotation and scale invariant properties. However, it suffers from nonuniform sampling which makes it not suitable for applications in which the registered images are altered or occluded. Inspired by LPT, this paper presents a new registration algorithm that addresses the problems of the conventional LPT while maintaining the robustness to scale and rotation. We introduce a novel adaptive polar transform (APT) technique that evenly and effectively samples the image in the Cartesian coordinates. Combining APT with an innovative projection transform along with a matching mechanism, the proposed method yields less computational load and more accurate registration than that of the conventional LPT. Translation between the registered images is recovered with the new search scheme using Gabor feature extraction to accelerate the localization procedure. Moreover an image comparison scheme is proposed for locating the area where the image pairs differ. Experiments on real images demonstrate the effectiveness and robustness of the proposed approach for registering images that are subjected to occlusion and alteration in addition to scale, rotation, and translation.     

基于角度分类的快速图像检索系统

基于内容的图像检索是近年来计算机视觉领域的重要方向之一,如何快速准确地匹配视觉信息内容是图像检索最关键的部分。目前大多数检索方法采用BOF(bag of features)算法,该算法的检索精度较低,且运行速度较慢。提出了一种新的匹配方法,提高检索精度的同时有效减少了检索时间。本算法利用特征点的四个相对独立的角度对其进行分类,可大幅减少需要比较的特征算子的数量,并对每一分类中的特征点使用k-means算法聚类,得到若干个聚类中心。本方法对每一聚类的特征点进行汉明编码,并采用倒排表的方式进行信息存储。实验对象使用Holiday图像库,结果显示,检索精度和检索速度较原先算法得到了较大程度的改善,检索精度最高可提高55.9%,至0.8557,检索时间最多可降低49.3%,至0.35s。