Directional variation adaptive image resolution enhancement

The goal of image interpolation is to produce a high-resolution image from its low-resolution counterpart. It has significant applications in video sensor network, where the resolution of images usually needs to be enhanced at the end user due to the limited transmission bandwidth. The key challenge of image interpolation is to preserve the edge structure of the image. In this paper, a new image interpolation approach is proposed to adaptively adjust the interpolation according to the directional variations of images. More specifically, at each pixel position to be interpolated, its neighboring pixels are projected onto 1D direction according to a number of proposed patterns. Then the direction, of which the variation is smallest, is chosen as the direction to perform image interpolation. Experimental results are provided to show that the proposed approach outperforms several conventional edge-directed image interpolation algorithms.     

Fast single image super-resolution using estimated low-frequency k-space data in MRI

PurposeSingle image super-resolution (SR) is highly desired in many fields but obtaining it is often technically limited in practice. The purpose of this study was to propose a simple, rapid and robust single image SR method in magnetic resonance (MR) imaging (MRI).MethodsThe idea is based on the mathematical formulation of the intrinsic link in k-space between a given (modulus) low-resolution (LR) image and the desired SR image. The method consists of two steps: 1) estimating the low-frequency k-space data of the desired SR image from a single LR image; 2) reconstructing the SR image using the estimated low-frequency and zero-filled high-frequency k-space data. The method was evaluated on digital phantom images, physical phantom MR images and real brain MR images, and compared with existing SR methods.ResultsThe proposed SR method exhibited a good robustness by reaching a clearly higher PSNR (25.77dB) and SSIM (0.991) averaged over different noise levels in comparison with existing edge-guided nonlinear interpolation (EGNI) (PSNR=23.78dB, SSIM=0.983), zero-filling (ZF) (PSNR=24.09dB, SSIM=0.985) and total variation (TV) (PSNR=24.54dB, SSIM=0.987) methods while presenting the same order of computation time as the ZF method but being much faster than the EGNI or TV method. The average PSNR or SSIM over different slice images of the proposed method (PSNR=26.33 dB or SSIM=0.955) was also higher than the EGNI (PSNR=25.07dB or SSIM=0.952), ZF (PSNR=24.97dB or SSIM=0.950) and TV (PSNR=25.70dB or SSIM=0.953) methods, demonstrating its good robustness to variation in anatomical structure of the images. Meanwhile, the proposed method always produced less ringing artifacts than the ZF method, gave a clearer image than the EGNI method, and did not exhibit any blocking effect presented in the TV method. In addition, the proposed method yielded the highest spatial consistency in the inter-slice dimension among the four methods.ConclusionsThis study proposed a fast, robust and efficient single image SR method with high spatial consistency in the inter-slice dimension for clinical MR images by estimating the low-frequency k-space data of the desired SR image from a single spatial modulus LR image.     

双正交小波插值的CCD图像超分辨恢复

给出了一种构造双正交小波滤波器的设计方法,应用该方法得到的双正交小波对低分辨率图像进行小波分解。对分解后的图像结构进行分析,提出双三次插值与方向插值相结合的方法,对分解后图像插值。对插值后图像小波逆变换,得到超分辨率图像。应用于CCD图像,获得了比双线性插值高的峰值信噪比(25.3090dB),图像细节信息增强了。实验结果表明,算法应用于CCD图像不仅提高了图像的空间分辨率,同时也提高了图像的峰值信噪比,还比较好地保留了图像的边缘信息,且经过该方法恢复后的图像更适合人眼观察,细节丰富,更加清晰,畸变小,是一种提高CCD图像空间分辨率的有效方法。     

Improved matrix inversion in image plane parallel MRI

A new 3D parallel magnetic resonance imaging (MRI) method named Generalized Unaliasing Incorporating Support constraint and sensitivity Encoding (GUISE) is presented. GUISE allows direct image recovery from arbitrary Cartesian k-space trajectories. However, periodic k-space sampling patterns are considered for reconstruction efficiency. Image recovery methods such as 2D SENSE (SENSitivity Encoding) and 2D CAIPIRINHA (Controlled Aliasing In Parallel Imaging Results IN Higher Acceleration) are special instances of GUISE where specific restrictions are placed on the k-space sampling patterns used. It is shown that the sampling pattern has large impacts on the image reconstruction error due to noise. An efficient sampling pattern design method that incorporates prior knowledge of object support and coil sensitivity profile is proposed. It requires no experimental trials and could be used in clinical imaging. Comparison of the proposed sampling pattern design method with 2D SENSE and 2D CAIPIRINHA are made based on both simulation and experiment results. It is seen that this new adaptive sampling pattern design method results in a lower noise level in reconstructions due to better exploitation of the coil sensitivity variation and object support constraint. In addition, elimination of the non-object region from reconstruction potentially allows an acceleration factor higher than the number of receiver coils used.     

基于模糊隶属度的图像空间距离修正插值算法

为解决传统图像插值算法存在的边缘模糊和边缘锯齿,提出了一种基于像素点模糊隶属度的图像自适应插值方法.该方法首先根据图像的梯度与相角特性,确定像素点的模糊隶属度,再根据图像的局部不对称性在一维方向上修正插值点空间距离,并将一维修正结果转化到二维图像空间,最终将修正后的空间距离应用到传统双线性插值和双立方插值中.实验结果表明,该算法改善了图像的信噪比,有效抑制了边缘锯齿和边缘模糊的发生.     

IKWI-net: A cross-domain convolutional neural network for undersampled magnetic resonance image reconstruction

Magnetic resonance imaging (MRI) is widely used to get the information of anatomical structure and physiological function with the advantages of high resolution and non-invasive scanning. But the long acquisition time limits its application. To reduce the time consumption of MRI, compressed sensing (CS) theory has been proposed to reconstruct MRI images from undersampled k-space data. But conventional CS methods mostly use iterative methods that take lots of time. Recently, deep learning methods are proposed to achieve faster reconstruction, but most of them only pay attention to a single domain, such as the image domain or k-space. To take advantage of the feature representation in different domains, we propose a cross-domain method based on deep learning, which first uses convolutional neural networks (CNNs) in the image domain, k-space and wavelet domain simultaneously. The combined order of the three domains is also first studied in this work, which has a significant effect on reconstruction. The proposed IKWI-net achieves the best performance in various combinations, which utilizes CNNs in the image domain, k-space, wavelet domain and image domain sequentially. Compared with several deep learning methods, experiments show it also achieves mean improvements of 0.91 dB in peak signal-to-noise ratio (PSNR) and 0.005 in structural similarity (SSIM).     

3D image registration using a fast noniterative algorithm

This note describes the implementation of a three-dimensional (3D) registration algorithm, generalizing a previous 2D version [Alexander, Int J Imaging Systems and Technology 1999;10:242-57]. The algorithm solves an integrated form of linearized image matching equation over a set of 3D rectangular sub-volumes ('patches') in the image domain. This integrated form avoids numerical instabilities due to differentiation of a noisy image over a lattice, and in addition renders the algorithm robustness to noise. Registration is implemented by first convolving the unregistered images with a set of computationally fast [O(N)] filters, providing four bandpass images for each input image, and integrating the image matching equation over the given patch. Each filter and each patch together provide an independent set of constraints on the displacement field derived by solving a set of linear regression equations. Furthermore, the filters are implemented at a variety of spatial scales, enabling registration parameters at one scale to be used as an input approximation for deriving refined values of those parameters at a finer scale of resolution. This hierarchical procedure is necessary to avoid false matches occurring. Both downsampled and oversampled (undecimating) filtering is implemented. Although the former is computationally fast, it lacks the translation invariance of the latter. Oversampling is required for accurate interpolation that is used in intermediate stages of the algorithm to reconstruct the partially registered from the unregistered image. However, downsampling is useful, and computationally efficient, for preliminary stages of registration when large mismatches are present. The 3D registration algorithm was implemented using a 12-parameter affine model for the displacement: u(x) = Ax + b. Linear interpolation was used throughout. Accuracy and timing results for registering various multislice images, obtained by scanning a melon and human volunteers in various stationary positions, is described. The algorithm may be generalized to more general models of the displacement field, and is also well suited to parallel processing.     

亚像元几何图像成像与超分辨恢复算法

随着光学成像到光电数字成像的转变,如何提高CCD的几何分辨率已成为研制高分辨光电成像系统亟待解决的问题。从研究现状入手,给出了现有算法并指出不足之处`,建立了亚像元超分辨成像数学模型,提出了亚像元的CCD几何超分辨方法:将两片线阵CCD集成在同一器件中,在线阵方向上错开半个像元,同时读出时间减半,最终交织重组图像数据,合成高分辨率图像。利用MATLAB软件对双线性插值方法及亚像元成像方法进行了仿真,并定性定量地分析了两种方法的效果。结果表明:亚像元方法合成图像分辨率约为低分辨率图像的2倍,且两组仿真图像中的峰值信噪比比双线性插值图像分别高出1.4864dB和2.2070dB,该方法可以显著地减轻欠采样引起的图像模糊,且实时性优于双线性插值方法。     

Computational integral imaging reconstruction method of 3D images using pixel-to-pixel mapping and image interpolation

In this paper, we propose a novel computational integral imaging reconstruction (CIIR) method to improve the visual quality of the reconstructed images using a pixel-to-pixel mapping and an interpolation technique. Since an elemental image is magnified inversely through the corresponding pinhole and mapped on the reconstruction output plane based on pinhole-array model in the conventional CIIR method, the visual quality of reconstructed output image (ROI) degrades due to the interference problem between adjacent pixels during the superposition of the magnified elemental images. To avoid this problem, the proposed CIIR method generates dot-pattern ROIs using a pixel-to-pixel mapping and substitutes interpolated values for the empty pixels within the dot-pattern ROIs using an interpolation technique. The interpolated ROIs provides a much improved visual quality compared with the conventional method because of the exact regeneration of pixel positions sampled in the pickup process without interference between pixels. Moreover, it can enable us to reduce a computational cost by eliminating the magnification process used in the conventional CIIR. To confirm the feasibility of the proposed system, some experiments are carried out and the results are presented.     

Estimation of mutual information objective function based on Fourier shift theorem: an application to eddy current distortion correction in diffusion tensor imaging

Diffusion tensor imaging requires correction of eddy current distortion in diffusion-weighted images. An effective retrospective correction approach is to transform a diffusion-weighted image to maximize the mutual information (MI) between the transformed diffusion-weighted image and the corresponding T2-weighted image. In the literature, either linear interpolation or partial volume interpolation is applied to estimate the MI objective function. However, these interpolation methods induce artifacts to the MI objective function, thus compromising correction results. In this work, the MI objective function is estimated based on interpolation using Fourier shift theorem. This method eliminates the artifacts incurred with the aforementioned interpolation methods. The algorithm is further improved by approximating pixel values using their nearest neighbors in the up-sampled spatial domain, resulting in dramatically increased computational efficiency without compromising the correction results. The effects of varying the number of quantization levels and using Parzen window filtering to smooth the MI objective function are also investigated to obtain optimized algorithm parameters. The diffusion tensor image quality after applying the proposed distortion correction method is significantly improved visually.     

A 3D trajectory for undersampling k-space in MRSI applications

Magnetic resonance spectroscopic imaging (MRSI) is a noninvasive technique for producing spatially localized spectra. MRSI presents the important challenge of reducing the scan time while maintaining the spatial resolution. The preferred approach for this is to use time-varying readout gradients to collect the spatial and chemical-shift information. Fast, three-dimensional (3D) spatial encoded methods also reduce the scan time. Despite the existence of several new and faster 3D encoded methods, or k-space trajectories, for magnetic resonance imaging (MRI), only stack of spirals and echo planar have been studied in 3D MRSI. A novel formulation for designing fast, 3D k-space trajectory applicable to 3D MRSI is presented. This approach is simple and consists of rays expanding from the origin of k-space into a revolving sphere, collecting spectral data of all 3D spatial k-space at different times in the same scan. This article describes this new method and presents some results of its application to 3D MRSI. This technique allows some degree of undersampling; hence, it is possible to reconstruct high-quality undersampled spectroscopic imaging in order to recognize different compounds in short scan times. Additionally, the method is tested in regular 3D MRI. This proposed method can also be used for dynamic undersampled imaging.     

Diffusion tensor image up-sampling: a registration-based approach

Diffusion weighted images (DWI), from which the corresponding diffusion tensor images (DTI) are estimated, are commonly acquired with anisotropic discretizations. Traditional methods to up-sample diffusion weighted images generally rely on scene-based interpolation and do not exploit structural information from the images. In this study, a DTI up-sampling framework is presented that incorporates the underlying anatomical shape information by means of non-rigid inter-slice registration. A strategy is proposed to reorient the interpolated tensor in order to maintain its proper orientation. Tests on phantom as well as on real data sets show that the proposed method is able to produce better results compared to scene based interpolation methods in terms of the accuracy of DWI/DTI interpolation, especially when diffusion tensor orientation is taken into account.     

改进Canny算法的红外人脸图像边缘轮廓提取

红外人脸图像的边缘轮廓特征对于红外人脸检测、识别等相关应用具有重要价值。针对红外人脸图像边缘轮廓提取时存在伪边缘的问题,提出了一种改进Canny算法的红外人脸图像边缘轮廓提取方法。首先通过对引导滤波算法引入“动态阈值约束因子”替换原始算法中的高斯滤波,解决了原始算法滤波处理不均匀和造成红外人脸图像弱边缘特征丢失的弊端;接着对原始算法的非极大值抑制进行了改进,在原始计算梯度方向的基础上又增加了4个梯度方向,使得非极大值抑制的插值较原始算法更加精细;最后改进OTSU(大津)算法,构造灰度-梯度映射函数确定最佳阈值,解决了原始算法人为经验确定阈值的局限性。实验结果表明:提出的改进Canny算法的红外人脸轮廓提取方法滤波后的图像,相较于原始Canny算法滤波处理,信噪比性能提升了34.40%,结构相似度性能提升了21.66%;最终的红外人脸边缘轮廓提取实验的优质系数值高于对比实验的其他方法,证明改进后的算法对于红外人脸图像边缘轮廓提取具有优越性。     

Increasing resolution of digital images using edge-based approach

This paper presents the results of our research aimed at obtaining new methods for increasing resolution of digital images giving better visual results. Classification of these methods and implementation of some algorithms is also shortly presented here. The main part of the paper presents modification and development of the new methods. Main new feature, which we added to the interpolation algorithms, consists in taking into consideration in the final image (i.e., after interpolation at high resolution) the edges detected in the original image (i.e., before interpolation). Summing an image and its properly processed edges in the interpolation process enables us to get final image characterized by better sharpness with simultaneous precise presentation of the image details interpolated to higher resolution.     

Shortcut in DIC error assessment induced by image interpolation used for subpixel shifting

In order to characterize errors of Digital Image Correlation (DIC) algorithms, sets of virtual images are often generated from a reference image by in-plane sub-pixel translations. This leads to the determination of the well-known S-shaped bias error curves and their corresponding random error curves. As images are usually shifted by using interpolation schemes similar to those used in DIC algorithms, the question of the possible bias in the quantification of measurement uncertainties of DIC softwares is raised and constitutes the main problematic of this paper. In this collaborative work, synthetic numerically shifted images are built from two methods: one based on interpolations of the reference image and the other based on the transformation of an analytic texture function. Images are analyzed using an in-house subset-based DIC software and results are compared and discussed. The effect of image noise is also highlighted. The main result is that the a priori choices to numerically shift the reference image modify DIC results and may lead to wrong conclusions in terms of DIC error assessment.     

A correction algorithm for undersampled images using dynamic segmentation and entropy based focus criterion

A post-processing technique is presented for correcting images undersampled in k-space. The method works by taking advantage of the image's background zeros (dynamically segmented through the application of a threshold) to extrapolate the missing k-space samples. The algorithm can produce good quality images from a small set of k-space frequencies with only a few iterations of simple matrix operations, using the image entropy as the focus criterion. It does not require any special patient preparation, extra pulse sequences, complex gradient programming or specialized hardware. This makes it a good candidate for any application that requires short scan times or where only few frequencies can be sampled.     

MR image reconstruction of sparsely sampled 3D k-space data by projection-onto-convex sets

In many rapid three-dimensional (3D) magnetic resonance (MR) imaging applications, such as when following a contrast bolus in the vasculature using a moving table technique, the desired k-space data cannot be fully acquired due to scan time limitations. One solution to this problem is to sparsely sample the data space. Typically, the central zone of k-space is fully sampled, but the peripheral zone is partially sampled. We have experimentally evaluated the application of the projection-onto-convex sets (POCS) and zero-filling (ZF) algorithms for the reconstruction of sparsely sampled 3D k-space data. Both a subjective assessment (by direct image visualization) and an objective analysis [using standard image quality parameters such as global and local performance error and signal-to-noise ratio (SNR)] were employed. Compared to ZF, the POCS algorithm was found to be a powerful and robust method for reconstructing images from sparsely sampled 3D k-space data, a practical strategy for greatly reducing scan time. The POCS algorithm reconstructed a faithful representation of the true image and improved image quality with regard to global and local performance error, with respect to the ZF images. SNR, however, was superior to ZF only when more than 20% of the data were sparsely sampled. POCS-based methods show potential for reconstructing fast 3D MR images obtained by sparse sampling.     

Volume reconstruction of freehand three-dimensional ultrasound using median filters

Objectives

This paper aims to apply median filters for reducing interpolation error and improving the quality of 3D images in a freehand 3D ultrasound (US) system.

Background and motivation

Freehand 3D US imaging has been playing an important role in obtaining the entire 3D impression of tissues and organs. Reconstructing a sequence of irregularly located 2D US images (B-scans) into a 3D data set is one of the key procedures for visualization and data analysis.

Methods

In this study, we investigated the feasibility of using median filters for the reconstruction of 3D images in a freehand 3D US system. The B-scans were collected using a 7.5 MHz ultrasound probe. Four algorithms including the standard median (SM), Gaussian weighted median (GWM) and two types of distance-weighted median (DWM) filters were proposed to filter noises and compute voxel intensities. Qualitative and quantitative comparisons were made among the results of different methods based on the image set captured in freehand from the forearm of a healthy subject. A leave-one-out approach was used to demonstrate the performance of the median filters for predicting the removed B-scan pixels.

Results

Compared with the voxel nearest-neighbourhood (VNN) and distance-weighted (DW) interpolation methods, the four median filters reduced the interpolation error by 8.0-24.0% and 1.2-21.8%, respectively, when 1/4 to 5 B-scans was removed from the raw B-scan sequence.

Conclusions

In summary, the median filters can improve the quality of volume reconstruction by reducing the interpolation errors and facilitate the following image analyses in clinical applications.     

光电混合相关识别系统中匹配滤波器缩放比例的确定方法

在光电混合匹配滤波相关识别系统中,由于输入图像的频谱与匹配滤波器的形成方式不同,不能直接实现对应频率成份的准确对准。这个问题需要通过对匹配滤波器进行缩放来解决。根据透镜的傅里叶变换性质和计算全息理论,分别推导出输入图像的频谱的表达式和滤波空间光调制器加载了匹配滤波器后的透射函数的表达式,并根据这两个表达式得到匹配滤波器的缩放比例的公式,然后在实际的光学识别系统中对根据该公式计算的缩放比例进行了验证。实验结果表明,在该缩放比例下,相关峰的质量有较大改善,这说明此时输入图像的频谱与匹配滤波器对应频率成份实现了较好地对准。