基于流形学习的人脸图像超分辨率技术研究

超分辨率技术是由低分辨率图像复原出高分辨率图像的技术。针对人脸图像进行基于学习的超分辨率技术研究,将流形学习算法融入到超分辨率算法中,并且将其用于人脸图像的超分辨率复原。对流形学习应用于基于学习的超分辨率原理进行了介绍。为了使得人脸图像取得更好的复原效果,对特征提取模板进行改进,使得新的特征提取模板考虑更多的像素之间的相关性,并更好地抑制噪声的影响,保留了更多的特征信息。加入了新的特征(即拉普拉斯特征)。该特征突出的边缘细节,保持了人脸图像鲜明的轮廓和清晰的边缘信息。实验结果表明,算法复原出的人脸图像更接近于真实图像,具有更高的峰值信噪比。     

稀疏阈值的超分辨率图像重建

为了解决基于字典学习的超分辨重构算法耗时过长的问题,提出了基于稀疏阈值模型的图像超分辨率重建方法。首先,将联合字典理论与图像块稀疏阈值方法相结合,训练得到高、低分辨率过完备图像字典对。接着,通过稀疏阈值OMP算法对图像特征块进行稀疏表示。然后,通过高分辨率字典重构出初始的超分辨图像。最后,通过改进迭代反投影算法对初始的超分辨图像进行全局优化,从而进一步提高图像重构质量。实验结果表明,超分辨图像重构平均峰值信噪比(PSNR)为30.1 d B,平均结构自相似度(SSIM)为0.937 9,平均计算时间为10.2 s。有效提高了超分辨重构的速度,改善了重构高分辨图像的质量。     

基于双字典和稀疏表示的医学图像超分辨率重建

在医学影像图像处理过程中,由于成像技术和成像时间的限制,还无法获取满足诊断需求的清晰图像,这使得在现有技术和极短时间内所获取的医学病理图像需要进行超分辨率的重建处理;基于学习的图像超分辨率思想是从已建立的先验模型中重建出高频细节;在文章中,将要估计的高频信息认为是由主要高频和冗余高频两部分组成,提出了一种基于双字典学习和稀疏表示的医学图像超分辨率重建算法,由主要字典学习和冗余字典学习组成,分别渐近地恢复出主要高频细节和冗余高频细节;实验结果的数据分析和视觉效果显示,所提出双层递进方法能够恢复更多的图像细节且在性能指标上比现有的其他几种方法均有所提高。     

基于ZedBoard的超分辨率视频复原系统

为了实现超分辨率视频图像的实时复原,设计了以ZedBoard可编程片上系统为基础的超分辨率视频复原系统。系统包括基于V4L2(Video for Linux 2)的USB摄像头视频采集、基于小波变换的超分辨率复原算法处理和基于Qt的图形用户界面制作以及视频输出。采用双线性、双立方和小波变换算法分别对Lena图像进行复原处理,峰值信噪比PSNR值分别为29.516、29.843、31.368。实验结果表明,提出的基于小波的超分辨率复原算法优于传统的插值算法,基于ZedBoard的超分辨视频复原系统复原效果良好。     

多尺度融合超分辨率算法在无人机探测中的应用

基于光电传感器的低慢小无人机探测系统能够快速准确地发现并识别无人机目标,但远距离非合作无人机目标在图像中像素比重过小,特征退化较明显,使识别率大大降低.图像超分辨技术能够从低分辨率目标图像区域中获得高分辨率图像并恢复更多的细节特征,现有超分辨技术很难在保证推理速度的前提下兼容图像的高低频特征,因此为了满足探测系统的需求...     

基于GoogLeNet和空间谱变换的高光谱图像超分辨率方法

为了提高高光谱图像的空间分辨率,提出了一种基于GoogLeNet和空间谱变换的高光谱图像超分辨率(SR)方法.设计出遥感图像的光谱SR框架,对图像中不同反射光谱进行提取;采用GoogLeNet的稀疏编码对粗像素光谱进行放大,并投影到高分辨率字典上,将潜在SR表示进行反转,以获得超分辨光谱;为了提高图像重构的保真度,利用...     

基于自适应正则化的超分辨率重建方法

传统正则化超分辨重建得到的图像往往存在过度平滑或伪信息残留的问题,结合超分辨重建模型对重建图像伪信息的产生进行了分析,针对传统方法的不足提出了基于图像区域信息自适应的正则化方法,通过图像的区域信息将图像划分为平滑区与非平滑区域,对不同区域选用不同的先验模型进行约束。同时考虑人眼的视觉感知特性,结合区域信息实现正则化参数的自适应选取。实验结果表明该方法在抑制重建图像伪信息的同时能有效保护细节,效果要优于传统方法与单一的先验模型约束,对于红外与可见光图像重建效果的提升提供了一定的理论参考。     

Learning-based compressed sensing for infrared image super resolution

This paper presents an infrared image super-resolution method based on compressed sensing (CS). First, the reconstruction model under the CS framework is established and a Toeplitz matrix is selected as the sensing matrix. Compared with traditional learning-based methods, the proposed method uses a set of sub-dictionaries instead of two coupled dictionaries to recover high resolution (HR) images. And Toeplitz sensing matrix allows the proposed method time-efficient. Second, all training samples are divided into several feature spaces by using the proposed adaptive k-means classification method, which is more accurate than the standard k-means method. On the basis of this approach, a complex nonlinear mapping from the HR space to low resolution (LR) space can be converted into several compact linear mappings. Finally, the relationships between HR and LR image patches can be obtained by multi-sub-dictionaries and HR infrared images are reconstructed by the input LR images and multi-sub-dictionaries. The experimental results show that the proposed method is quantitatively and qualitatively more effective than other state-of-the-art methods.     

基于POCS-MPMAP合成算法的超分辨率图像复原

阐述了凸集投影理论(POCS)和基于Markov约束的Poisson-MAP超分辨率图像复原算法(MPMAP).为了能获得更大的带宽扩展和更好的复原图像,把POCS理论与MPMAP算法结合在一起,提出基于POCS-MPMAP的合成超分辨率图像复原算法.由此获得了改善的图像复原结果.     

卷积稀疏表示图像融合与超分辨率联合实现

为避免图像融合与超分辨率分步实现的不足,提出了基于卷积稀疏表示的融合与超分辨率重建联合实现方法。假设低分辨率与高分辨率图像之间具有相同的稀疏特征图,设计了一种高、低分辨率滤波器联合学习框架,实现对图像高低频成分的分离,并根据不同成分的形态特性设计了不同的融合规则:对于高频成分,根据稀疏特征图亮度信息和像素活跃性水平,设计了一种像素显著性度量方案来指导高频特征图的融合;对于低频成分,根据脉冲耦合神经网络能捕获邻域相似像素点火的特性,设计了低频成分融合方法。所提方法不需要将图像分割成重叠的块,避免块向量化的缺陷。实验结果表明,能有效提高图像融合的质量。     

基于亚像素扫描的超分辨技术在高分辨X射线显微镜中的应用

在不改变现有硬件条件的情况下,开展超分辨扫描重建方法,可以在不增加系统成本的基础上提高高分辨X射线显微镜的成像性能.设计了基于亚像素扫描的超分辨扫描模式,按照设计的调制方式进行亚像素位移的移动,采集多幅具有互补信息的低分辨率图像;然后基于系统的点扩散函数,对高分辨率图像进行复原;最后结合POCS超分辨重建算法重建出高分辨图像.实验结果表明,10倍光耦探测器下的衬度噪声比提高了20%左右,空间分辨力提高了0.2μm(约15%),细节分辨能力超过探测器像素尺寸1.35μm的限制,可以看到在低分辨率图像中看不到的细节.实验说明用超分辨技术提高高分辨X射线显微镜的分辨率是有意义的.     

Experimental investigation of the quality of lensless super-resolution ghost imaging via sparsity constraints

Ghost imaging via sparsity constraints (GISC) can nonlocally realize super-resolution imaging. Factors influencing the quality of lensless super-resolution GISC are investigated and the experimental results show that, the quality of GISC is enhanced as the object?s sparse ratio in the representation basis or the spatial transverse coherence lengths on the object plane are decreased. The differences between ghost imaging (GI) and GISC are also discussed.     

Fusion method for infrared and visible images by using non-negative sparse representation

In this paper, an interesting fusion method, named as NNSP, is developed for infrared and visible image fusion, where non-negative sparse representation is used to extract the features of source images. The characteristics of non-negative sparse representation coefficients are described according to their activity levels and sparseness levels. Multiple methods are developed to detect the salient features of the source images, which include the target and contour features in the infrared images and the texture features in the visible images. The regional consistency rule is proposed to obtain the fusion guide vector for determining the fused image automatically, where the features of the source images are seamlessly integrated into the fused image. Compared with the classical and state-of-the-art methods, our experimental results have indicated that our NNSP method has better fusion performance in both noiseless and noisy situations.     

Infrared super-resolution imaging method based on retina micro-motion

With the wide application of infrared focal plane arrays (IRFPA), military, aerospace, public security and other applications have higher and higher requirements on the spatial resolution of infrared images. However, traditional super-resolution imaging methods have increasingly unable to meet this requirement in technology. In this paper, we adopt the achievement that the human retina micro-motion is the important reason why the human has the hyperacuity ability. Based on the achievement, we bring forward an infrared super-resolution imaging method based on retina micro-motion. In the method, we use the piezoelectric ceramic equipment to control the infrared detector moving variably within a plane parallel to the focal plane. The motion direction is toward each other into a direction of 90°. In the four directions of the movement, we get four sub-images and generate a high spatial resolution infrared image by image interpolation method. In the process of the shifting movement of the detector, we set the threshold of the detector response and record the response time difference when adjacent pixel responses are up to the threshold. By the method, we get the object’s edges, enhance them in the high resolution infrared image and get the super-resolution infrared image. The experimental results show that our proposed super-resolution imaging methods can improve the spatial resolution of the infrared image effectively. The method will offer a new idea for the super-resolution reconstruction of infrared images.     

基于双阈值Huber范数估计的图像正则化超分辨率算法

多帧图像超分辨率算法利用图像间的互补信息, 可以从一系列具有亚像素位移的低分辨率影像数据中重建出高分辨率图像. 在众多超分辨率算法中, 正则化方法以其求解病态问题的有效性而被广泛应用, 但在此类方法中, 最优估计算子的估计准确度对最后的重建结果有着较大的影响. 本文在现有正则化超分辨率重建算法的基础上, 提出了一种基于双阈值Huber范数的极大似然估计算子, 可以提高Huber范数对于阈值取值的容忍性和算子估计精度; 并给出了基于该算子的正则化超分辨率算法的迭代公式. 通过对仿真图像进行重建, 结果表明算法可有效地抑制各种噪声并保证重建效果; 同时将此算法应用于实际图像的超分辨率重建, 有效地提高了目标影像的空间分辨率. 关键词： 图像处理 图像重建 超分辨率 正则化     

应用稀疏表示的宽带相干超分辨方位估计方法

针对无源声呐中宽带相干处理对邻近弱目标分辨能力仍不足的问题,提出一种相干聚焦后稀疏空间谱求解的宽带超分辨方位估计方法。首先将不同频点数据相干聚焦到参考频点,构成多测量矢量稀疏求解问题,然后采用二阶锥规划内点算法进行稀疏空间谱拟合,实现了宽带弱目标超分辨方位估计。仿真试验结果表明,该方法在弱目标分辨和方位估计精度方面优于宽带非相干和相干信号子空间高分辨处理方法。32元拖曳阵海试数据的处理结果验证了该方法的优越性能,同时说明了该方法在无源声呐探测中应用的可行性。      

Texton-based super-resolution for achieving high spatiotemporal resolution in hybrid camera system

Many super-resolution methods have been proposed to enhance the spatial resolution of images by using iteration and multiple input images. In a previous paper, we proposed the example-based super-resolution method to enhance an image through pixel-based texton substitution to reduce the computational cost. In this method, however, we only considered the enhancement of a texture image. In this study, we modified this texton substitution method for a hybrid camera to reduce the required bandwidth of a high-resolution video camera. We applied our algorithm to pairs of high- and low-spatiotemporal-resolution videos, which were synthesized to simulate a hybrid camera. The result showed that the fine detail of the low-resolution video can be reproduced compared with bicubic interpolation and the required bandwidth could be reduced to about 1/5 in a video camera. It was also shown that the peak signal-to-noise ratios (PSNRs) of the images improved by about 6 dB in a trained frame and by 1.0–1.5 dB in a test frame, as determined by comparison with the processed image using bicubic interpolation, and the average PSNRs were higher than those obtained by the well-known Freeman’s patch-based super-resolution method. Compared with that of the Freeman’s patch-based super-resolution method, the computational time of our method was reduced to almost 1/10.     

An adaptive super-resolution method based on regional pixel information and ringing artifacts suppression

Multi-frame image super-resolution (SR) aims to utilize information from a set of low-resolution (LR) images to compose a high-resolution (HR) one. In this paper, a novel multi-frame image super-resolution algorithm is proposed based on regional pixel information and ringing artifacts suppression. Firstly, a new regularization term which adopts Regional Adaptive Weight Coefficients (RAWC) is produced to keep edges and flat regions. After detailed analysis, an iterative process is given for image reconstruction. Then an adaptive term according to the local variance of iterative correction image is designed to evaluate the ringing artifacts. Finally, the original iteration is updated by adding the restraint term for better visual effects and lower noise of reconstructive HR image. Thorough experimental results show the proposed algorithm is effective for SR reconstruction and ringing artifacts suppression.     

平场复用多焦点结构光照明超分辨显微成像

多焦点结构光照明显微技术(multifocal structured illumination microscopy, MSIM)能在50μm的成像深度内和1Hz的成像速度下实现两倍于衍射极限分辨率的提升,相比传统的宽场结构光照明显微技术,具有较大的成像深度和层析能力,更适合应用于厚样品的长时程三维超分辨成像.然而, MSIM存在成像速度慢、图像处理过程复杂等问题.本文提出了一种基于平场复用多焦点结构光照明的快速超分辨显微成像方法和系统(flat-field multiplexed MSIM, FM-MSIM),通过在照明光路中插入光束整形器件,将高斯光束转变为均为分布的平顶光束,提高激发点阵的强度均匀性和扩大视场;通过将每个衍射受限的激发点沿y方向延长,形成新的多路复用多焦点阵照明图案,提高能量利用率,减少扫描步数,进而提高成像速度和信噪比;结合基于多重测量矢量模型的稀疏贝叶斯学习图像重构算法,简化图像重构步骤,在保证空间分辨率的同时实现至少4倍于传统MSIM的成像速度.在此基础上,利用搭建的FM-MSIM系统进行了BSC细胞微管样片和小鼠肾切片标准样片的超分辨成像实验,实验结果证明...     

基于微分图像自相关的离焦模糊图像盲复原

针对离焦模糊图像的盲复原算法的研究具有重要的实际意义和实用价值.根据光学离焦成像模型,研究提出了一种基于微分图像自相关的离焦模糊图像超分辨力盲复原算法,即首先采用拉普拉斯算子对离焦模糊图像进行二阶微分并求微分图像的自相关,然后从自相关结果所包含的信息中确定离焦模糊半径,最后以离焦模糊模型结合MPMAP超分辨力复原算法对离焦模糊图像进行肓复原.实验证明:算法能够以较高的精度估计出离焦模糊半径并实现离焦模糊图像的盲复原,该算法较其它同类算法在减少计算过程中需要考虑的各类因素的同时也减少了计算量,提高了结果精度,依靠超分辨力复原算法获取更多的复原图像信息,已在实际刑侦和物证鉴定的离焦模糊图像判读和鉴定中获得成功应用.