基于粒子滤波的人脸图像超分辨率重建方法

将人脸图像超分辨率重建描述为人脸混合模型的纹理和位置参数的贝叶斯概率估计问题,将超分辨率重建的图像配准和像素融合这两个过程置于统一的概率估计框架下,并利用基于粒子滤波的参数估计算法,同时估计纹理和位置参数,从而实现人脸图像的超分辨率重建.包含灰度和位置两种先验信息的人脸混合模型,同时用于超分辨率重建的两个过程中,提高了图像配准精度和重建算法的性能,避免了通常方法在获得准确鲁棒的运动场估计时需要清晰的高分辨图像,而获得清晰的高分辨图像时又需要准确鲁棒运动场估计的困境.正面人脸合成序列图像实验结果表明,该方法获得的重建结果较为理想.     

基于非线性最小二乘的图像自适应SR重建以及运动估计

图像超分辨率（SR）重建是利用数字信号处理技术由一系列低分辨率观测图像得到高分辨率图像。为了扩展SR技术的应用范围,提出了一种同时进行图像超分辨率重建和全局运动估计的方法。该方法首先基于最大后验概率(MAP)给出了图像SR重建和运动估计框架,该框架不仅考虑了前后两次迭代所得的HR图像差值对最终重建图像的影响,而且引入了不同LR图像对重建图像的重要性权值,使得算法具有自适应性;然后将总体框架转换为图像SR重建模型和运动估计模型;最后基于非线性最小二乘法对模型进行优化求解,得出了SR重建图像及其全局运动域。实验表明,该方法不仅图像重建效果良好,并有着良好的收敛性。     

基于高斯金字塔的图像运动估计算法

运动估计是图像超分辨率重建中的关键环节,直接影响超分辨重建的结果质量。为减少运动估计搜索点数,提高搜索速度,提出一种基于高斯金字塔分层思想的小十字形搜索算法。算法把图像构建成一个两层高斯金字塔,在上层使用小十字形搜索算法估计初始运动矢量,并通过提前终止策略来提前结束搜索;在下层以上层估计结果作为初始值,估计最终图像运动矢量。在标准图像序列上进行的实验结果表明,该算法在保持搜索精度的前提下能明显提高搜索速度;特别对于在运动偏差较大的情况下,提高效果更显著。     

基于迭代反投影的超分辨率图像重建

提出了一种结合频域运动估计和迭代反投影的超分辨率图像重建算法。根据输入低分辨率序列图像各帧之间的傅立叶变换相位差,估计出每幅低分辨率图像相对于参考低分辨率图像的子象素位移;依据所得的子象素位移并结合迭代反投影算法,实现了超分辨率图像重建。实验结果表明,该算法是一种有效的超分辨率图像重建方法。     

基于混合POCS的视频图像超分辨率重建

提出混合凸集投影算法HPOCS对视频图像进行超分辨率重建,利用连续视频图像间的互异信息生成更高分辨率的视频图像.该算法首先采用图像匹配估计视频图像间的运动位移;然后进行基于的APEX盲解卷积,估计点扩散函数和理想视频图像;最后在凸集投影的理论框架下进行图像重建.实验表明,HPOCS重建后,视频图像的分辨率相对于原始图像、双线性内插图像和POCS重建图像明显提高,图像边缘更加清晰,细节信息更加突出.     

基于加速稳健特征匹配和凸集投影算法的超分辨率重建

针对不理想的配准结果会导致超分辨率重建失败的问题,提出了一种基于加速健壮特征(SURF)匹配和凸集投影(POCS)的超分辨率重建算法。该算法首先采用SURF算法进行连续帧图像的配准,估计图像序列的运动位移;然后根据运动估计结果,在POCS理论框架下进行图像重建。实验结果表明,该方法能够较明显地改善图像的视觉效果,获得较丰富的细节信息,且具有较好的噪声抑制能力。     

基于LMS自适应算法的图像去模糊研究

传统单幅图像去模糊方法需要稀疏先验约束,导致计算量较大。为此,在自适应最小均方误差(LMS)算法的基础上,提出一种点扩散函数(PSF)估计方法。利用模糊图像得到有效突出边缘,作为自适应滤波器的输入信号,并将模糊图像作为滤波器的期望信号,用以估计PSF。在非盲去卷积过程中,采用各项异性正规化方法对清晰图像进行约束,以减少恢复图像的振铃效应。实验结果表明,该方法不需要先验约束,对运动和非运动模糊图像均可适用,在保留图像细节的同时能抑制平滑区域的噪声。     

改进块匹配与初始估计的POCS图像重建

POCS（凸集投影）算法进行图像超分辨率重建时,图像配准和初始估计对重建结果有重要影响。为解决传统块匹配算法小块匹配易受噪声干扰,大块匹配对运动对象跟踪不精确的问题,本文以空间十字频率法判断图像的平坦程度,根据图像平坦度进行匹配块的分割,实现了匹配块大小的动态选择;针对单帧初始估计包含信息量少的不足,将所有已知低分辨图像与插值图像配准,以最优插值法作用后的插值图像作为初始估计。实验证明,改进的配准算法消除了固定块配准方式在噪声和局部运动同时存在时的不足,且算法简单,运算量较小;新的初始估计生成方式则提高了初始估计的峰值信噪比,加快了重建的收敛速度。     

基于全局运动模型配准的图像超分辨重建

把全局运动模型配准算法运用到序列图像超分辨重建中,通过与优化的基于频域的配准法进行对比,在运动模型可以准确地反映物体运动状态的情况下,该算法能够更精确地估计运动参数,从而确保重建后的高分辨率图像拥有更多细节信息。同时,阐述了参与重建的低分辨率图像帧数越多,重建精度会越高,但随着帧数的增多,重建误差降低幅度会越低,而算法复杂度及其耗时会过多地增加,因此提出应根据对重建精度的要求而确定参与重建的低分辨率图像的帧数。     

一种视频编码中块效应噪声的估计和抑制策略

提出了一种视频编码块效应噪声的估计和抑制策略，它在编码器端估计块效应噪声，并将少量的估计结果随码流传送至解码端，再在解码端根据这一结果进行自适应邻域后处理。这种策略不仅使得块效应噪声估计的结果更为精确，而且削减了解码器端噪声估计算法的开销，使得解码器端的设计更为简单。本文提出的分块块效应噪声估计与自适应邻域后处理算法配合，不仅很好地抑制了编码产生振铃噪声和块效应噪声，而且完好地保持了图像的边缘。实验数据和实际应用效果都表明，本文提出的算法有效地提高了重建图像的主客观质量。     

Using pyramids to define local thresholds for blob detection

A method of detecting blobs in images is described. The method involves building a succession of lower resolution images and looking for spots in these images. A spot in a low resolution image corresponds to a distinguished compact region in a known position in the original image. Further, it is possible to calculate thresholds in the low resolution image, using very simple methods, and to apply those thresholds to the region of the original image corresponding to the spot. Examples are shown in which variations of the technique are applied to several images.     

大旋角无人机影像全自动拼接方法研究

低空无人机抗风能力弱、稳定性差,影像旋偏角大且存在突变,无法按照常规正射影像镶嵌方法获得全区域拼接影像。为此,提出一种基于尺度不变特征变换(SIFT)特征匹配与多分辨率样条融合的低空无人机影像全自动拼接方法。对非量测影像进行畸变校正,利用查找表设计多幅影像快速畸变校正算法。采用SIFT特征的单应约束影像匹配算法,计算相邻影像的最优变换矩阵。给出最优变换矩阵的多分辨率样条融合影像拼接算法。实验结果表明,该方法能够获得大量稳定的匹配点对,影像间几何变换关系稳定,得到的拼接影像无缝清晰,适用于大旋角、低稳定性的低空无人机影像非摄影测量快速拼接。     

Reversible secret-image sharing with high visual quality

This paper proposes a reversible secret-image sharing scheme for sharing a secret image among 2n shadow images with high visual quality (i.e., they are visually indistinguishable from their original images, respectively). In the proposed scheme, not only can the secret image be completely revealed, but the original cover images can also be losslessly recovered. A difference value between neighboring pixels in a secret image is shared by 2n pixels in 2n shadow images, respectively, where n?≥?1. A pair of shadow images which are constructed from the same cover image are called brother stego-images. To decrease pixel values changed in shadow images, each pair of brother stego-images is assigned a weighted factor when calculating difference values to be shared. A pixel in a cover image is recovered by calculating the average of corresponding pixels in its brother stego-images. A single stego-image reveals nothing and a pair of pixels in brother stego-images reveals partial difference value between neighboring secret pixels. The more brother stego-images are collected, the more information in the secret image will be revealed. Finally, a secret image will be completely revealed if all of its brother stego-images are collected.     

An automatic geometric precision correction system based on hierarchical registration for HJ-1 A/B CCD images

On 6 September 2008, two optical satellites, HJ-1 A and B (HJ-1 A/B), were successfully launched from China. However, the system geometric correction products of the HJ-1 A/B charge-coupled device (HJ-1 images) have low geometric precision and need to be corrected. The HJ-1 images have a large aspect angle, a wide swath width, and a large image size. Furthermore, the local geometric distortions are too complex in one scene. Given these characteristics of HJ-1 images, geometric correction is still a challenging work. This article proposes an automatic geometric precision correction system (GPCS) based on the automatic registration between HJ-1 images and Landsat Thematic Mapper images. First, the coarse image matching method based on geometric-restricted scale-invariant feature transform (SIFT) is used to determine the coarse global transformation between the HJ-1 image and the reference image. Second, inspired by the hierarchical method of non-rigid registration for medical images, a hierarchical image matching approach is proposed based on the combination of SIFT feature points and template matching. This approach decomposes a matching problem of a whole image into numerous matching problems of image blocks and can overcome the impact of local distortions in HJ-1 images. Hierarchical random sample consensus (RANSAC) based on digital elevation model (H-RANSAC) is used to remove incorrect control points. Third, an HJ-1 image is rectified using a triangulated irregular network. Finally, the automatic evaluation method based on automatic image matching between the corrected HJ-1 image and the reference image is adopted to evaluate the geometric precision. On the one hand, experiments on eight HJ-1 images demonstrate the efficiency and accuracy of the different steps of GPCS. On the other hand, experiments on 1000 HJ-1 images also demonstrated the robustness, accuracy, and suitability for batch processing.     

位平面和SIFT相结合的图像匹配方法

针对传统的相关匹配算法计算量大,对图像旋转敏感等问题,提出了一种位平面和尺度不变特征变换（SIFT）相结合的图像匹配算法。将待拼接的两幅图像[A、][B]各自分解为8个位平面,对两幅图像都选择前4个具有视觉信息的位平面[A1A2A3A4]和[B1B2B3B4];对[A1A2、][A2A3、][A3A4]图像进行异或运算,得到3幅图像。由于异或后的图像[A1A2]具有足够的细节部分,轮廓却不清晰,图像[A3A4]轮廓清晰,但是丢失了太多细节,而图像[A2A3]具有清晰的轮廓,又具有足够的细节信息,所以采用图像[A2A3],然后与原图像[A]进行异或得到[A],同时采用同样的方法得到图像[B],再次采用SIFT算法进行点对匹配,利用欧氏距离进行图像匹配,最后利用RANSAC进行图像容错处理,得到一幅匹配图像。实验结果表明,该算法有效地提高了匹配速度,对图像明暗变化、尺度旋转等具有较强的健壮性。     

Medical image retrieval based on unclean image bags

Traditional content-based image retrieval (CBIR) scheme with assumption of independent individual images in large-scale collections suffers from poor retrieval performance. In medical applications, images usually exist in the form of image bags and each bag includes multiple relevant images of the same perceptual meaning. In this paper, based on these natural image bags, we explore a new scheme to improve the performance of medical image retrieval. It is feasible and efficient to search the bag-based medical image collection by providing a query bag. However, there is a critical problem of noisy images which may present in image bags and severely affect the retrieval performance. A new three-stage solution is proposed to perform the retrieval and handle the noisy images. In stage 1, in order to alleviate the influence of noisy images, we associate each image in the image bags with a relevance degree. In stage 2, a novel similarity aggregation method is proposed to incorporate image relevance and feature importance into the similarity computation process. In stage 3, we obtain the final image relevance in an adaptive way which can consider both image bag similarity and individual image similarity. The experiments demonstrate that the proposed approach can improve the image retrieval performance significantly.     

基于光谱曲线小波系数分维测度的高光谱影像高效分割方法

针对高光谱遥感影像处理效率的问题,提出了一种基于高光谱曲线小波分解低频系数分维特征影像和高频系数分维特征影像相结合的高光谱遥感影像分割方法。对高光谱响应曲线的分形测度进行了分析,提出基于光谱曲线小波分解高频系数的分维算法,得到多尺度高光谱分形特征影像。设计了低频系数分维特征影像和高频系数分维特征影像相结合的高光谱影像分割算法。高光谱曲线小波系数分维特征影像分割实验结果表明:该算法可取得与光谱曲线直接分形测度特征影像分割一致结果,但效率优于直接分维特征影像分割。     

A Distributed Algorithm for Content Based Indexing of Images by Projections on Ritz Primary Images

Large collections of images can be indexed by their projections on a few “primary” images. The optimal primary images are the eigenvectors of a large covariance matrix. We address the problem of computing primary images when access to the images is expensive. This is the case when the images cannot be kept locally, but must be accessed through slow communication such as the Internet, or stored in a compressed form. A distributed algorithm that computes optimal approximations to the eigenvectors (known as Ritz vectors) in one pass through the image set is proposed. When iterated, the algorithm can recover the exact eigenvectors. The widely used SVD technique for computing the primary images of a small image set is a special case of the proposed algorithm. In applications to image libraries and learning, it is necessary to compute different primary images for several sub-categories of the image set. The proposed algorithm can compute these additional primary images “offline”, without the image data. Similar computation by other algorithms is impractical even when access to the images is inexpensive.     

图注意力网络的场景图到图像生成模型

目的 目前文本到图像的生成模型仅在具有单个对象的图像数据集上表现良好，当一幅图像涉及多个对象和关系时，生成的图像就会变得混乱。已有的解决方案是将文本描述转换为更能表示图像中场景关系的场景图结构，然后利用场景图生成图像，但是现有的场景图到图像的生成模型最终生成的图像不够清晰，对象细节不足。为此，提出一种基于图注意力网络的场景图到图像的生成模型，生成更高质量的图像。方法 模型由提取场景图特征的图注意力网络、合成场景布局的对象布局网络、将场景布局转换为生成图像的级联细化网络以及提高生成图像质量的鉴别器网络组成。图注意力网络将得到的具有更强表达能力的输出对象特征向量传递给改进的对象布局网络，合成更接近真实标签的场景布局。同时，提出使用特征匹配的方式计算图像损失，使得最终生成图像与真实图像在语义上更加相似。结果 通过在包含多个对象的COCO-Stuff图像数据集中训练模型生成64×64像素的图像，本文模型可以生成包含多个对象和关系的复杂场景图像，且生成图像的Inception Score为7.8左右，与原有的场景图到图像生成模型相比提高了0.5。结论 本文提出的基于图注意力网络的场景图到图像生成模型不仅可以生成包含多个对象和关系的复杂场景图像，而且生成图像质量更高，细节更清晰。     

Neural network based method for image halftoning and inverse halftoning

A hybrid neural network based method for halftoning and inverse halftoning of digital images is presented. The halftone image is performed by single-layer perceptron neural network (SLPNN), and its corresponding continuous-tone image is reconstructed by radial-basis function neural network (RBFNN). The combined training procedure produces halftone images and the corresponding continuous tone images at the same time. The PSNR performance and visual image quality of these contone images achieved is comparable to the well-known inverse halftoning methods. The resultant halftone images compared with the error diffusion halftone are visually good, too. Furthermore, we apply different kinds of halftone images to a bi-level image compression method, called Block Arithmetic Coding for Image Compression (BACIC), which is better than the current facsimile methods.