Speckle noise removal based on structural convolutional neural networks with feature fusion for medical image

Application of convolutional neural networks (CNNs) for image additive white Gaussian noise (AWGN) removal has attracted considerable attentions with the rapid development of deep learning in recent years. However, the work of image multiplicative speckle noise removal is rarely done. Moreover, most of the existing speckle noise removal algorithms are based on traditional methods with human priori knowledge, which means that the parameters of the algorithms need to be set manually. Nowadays, deep learning methods show clear advantages on image feature extraction. Multiplicative speckle noise is very common in real life images, especially in medical images. In this paper, a novel neural network structure is proposed to recover noisy images with speckle noise. Our proposed method mainly consists of three subnetworks. One network is rough clean image estimate subnetwork. Another is subnetwork of noise estimation. The last one is an information fusion network based on U-Net and several convolutional layers. Different from the existing speckle denoising model based on the statistics of images, the proposed network model can handle speckle denoising of different noise levels with an end-to-end trainable model. Extensive experimental results on several test datasets clearly demonstrate the superior performance of our proposed network over state-of-the-arts in terms of quantitative metrics and visual quality.     

基于双重规整化的气动退化图像盲复原算法

通过合理建立针对红外图像的气动光学效应退化模型，根据极大似然估计准则设计复原算法，同时针对随机变化的点扩展函数和噪声对目标图像尤其是图像中细节恢复的影响，对单一规整化方法进行了扩展，采用双重规整化策略，将规整化分为两个各有侧重点的层次来处理。在微机上进行了一些复原实验，并给出对比结果，证实了该算法的有效性。     

Sequential and parallel image restoration: neural networkimplementations

Sequential and parallel image restoration algorithms and their implementations on neural networks are proposed. For images degraded by linear blur and contaminated by additive white Gaussian noise, maximum a posteriori (MAP) estimation and regularization theory lead to the same high dimension convex optimization problem. The commonly adopted strategy (in using neural networks for image restoration) is to map the objective function of the optimization problem into the energy of a predefined network, taking advantage of its energy minimization properties. Departing from this approach, we propose neural implementations of iterative minimization algorithms which are first proved to converge. The developed schemes are based on modified Hopfield (1985) networks of graded elements, with both sequential and parallel updating schedules. An algorithm supported on a fully standard Hopfield network (binary elements and zero autoconnections) is also considered. Robustness with respect to finite numerical precision is studied, and examples with real images are presented.     

基于ResNet的红外与低照度可见光图像融合方法

针对传统红外与低照度可见光图像融合后,容易造成目标模糊不清、细节信息缺失等问题,本文提出一种低照度可见光图像预增强与残差网络(Residual Network, ResNet)相结合的图像融合方法。该方法首先利用单尺度Retinex(Single Scale Retinex, SSR)算法对低照度可见光图像进行增强预处理,得到增强的可见光图像。其次,利用ResNet-50分别从增强后的可见光图像和红外图像中提取深度特征。然后,采用L1范数对生成的深度特征进行正则化处理,并通过上采样操作将其分辨率恢复至输入图像大小,得到权重图。最后,使用加权平均策略获取融合图像。实验结果表明,本文算法能更好地保留输入图像的纹理细节和结构信息;使用TNO数据集与现有的三种典型算法对比,该算法融合结果的离散余弦特征互信息(FMI_dct)、小波特征互信息(FMI_w)、基于噪声评估的融合性能(Nabf)、结构相似度测量(SSIM)四种客观指标总体优于对比算法。     

基于PCNN图像分割的医学图像融合算法北大核心CSCD

为充分提取源图像间的互补信息,改进传统的图像融合算法在亮度维持、能量保留、边缘信息保持等方面的不足,本文提出了基于脉冲耦合神经网络(pulse coupled neural network, PCNN)图像分割的医学图像融合算法。该算法综合了非下采样剪切波变换(non-subsampled shearlet transform, NSST)与PCNN。首先,选取标准差较大的源图像作为被分割图像,标准差较小的源图像作为参照图像,将源图像进行NSST分解,获取源图像低频子带系数和高频子带系数;在低频融合中,利用参数自适应的PCNN对被分割图像的低频子带进行分割,根据分割结果获取融合低频子带系数;在高频融合中,采用以区域能量和与拉普拉斯能量和两者的乘积作为判断函数,获取融合高频子带系数;利用NSST逆变换获取融合图像。最后,应用本文提出的算法,对脑萎缩、急性中风和高血压性脑病等3组电脑断层扫描/磁共振成像(computerized tomography/magnetic resonance imaging, CT/MRI)图像进行了融合仿真,并将仿真结果与2018年后国际刊上提出的5种算法的融合图像进行比较。结果表明,应用本文提出的融合算法得到的图像,有效地增强了不同模态间的信息互补,保持了融合图像与源图像具有相同明亮程度,又保留了源图像低亮度部分的边缘信息,更加符合人眼视觉特性,具有更高的客观评价指标。     

基于卷积神经网络的红外图像融合算法

红外图像与可见光图像融合的目的是为人类观察或其他计算机视觉任务生成信息更加丰富的图像。本文针对深度学习近年来在计算机视觉领域取得的巨大成功,提出一种基于卷积神经网络的红外与可见光图像融合算法。首先,使用引导滤波和高斯滤波器组成的尺度感知边缘保护滤波器对输入的源图像进行多尺度分解,基础层利用像素强度分布的加权平均融合规则进行融合,细节层借助卷积神经网络对空间细节进行提取融合。实验结果表明,本文算法可以较好的将特定尺度信息进行保存,并减小滤波对边缘细节带来的光晕影响,融合后图像噪声较少,细节呈现的更加自然,并且适合人类视觉感知。     

Hopfield neural network based algorithms for image restoration andreconstruction. II. Performance analysis

For pt. I see ibid., vol.48, no.7, p.2105-18 (2000). In this paper, we analyze four typical sequential Hopfield (1982) neural network (HNN) based algorithms for image restoration and reconstruction, which are the modified HNN (PK) algorithm, the HNN (ZCVJ) algorithm with energy checking, the eliminating-highest-error (EHE) algorithm, and the simulated annealing (SA) algorithm. A new measure, the correct transition probability (CTP), is proposed for the performance of iterative algorithms and is used in this analysis. The CTP measures the correct transition probability for a neuron transition at a particular time and reveals the insight of the performance at each iteration. The general properties of the CTP are discussed. Derived are the CTP formulas of these four algorithms. The analysis shows that the EHE algorithm has the highest CTP in all conditions of the severity of blurring, the signal-to-noise ratio (SNR) of a blurred noisy image, and the regularization term. This confirms the result in many previous simulations that the EHE algorithms can converge to more accurate images with much fewer iterations, have much higher correct transition rates than other HNN algorithms, and suppress streaks in restored images. The analysis also shows that the CTPs of all these algorithms decrease with the severity of blurring, the severity of noise, and the degree of regularization, which also confirms the results in previous simulations. This in return suggests that the correct transition probability be a rational performance measure     

A regularized iterative image restoration algorithm

The development of the algorithm is based on a set theoretic approach to regularization. Deterministic and/or statistical information about the undistorted image and statistical information about the noise are directly incorporated into the iterative procedure. The restored image is the center of an ellipsoid bounding the intersection of two ellipsoids. The proposed algorithm, which has the constrained least squares algorithm as a special case, is extended into an adaptive iterative restoration algorithm. The spatial adaptivity is introduced to incorporate properties of the human visual system. Convergence of the proposed iterative algorithms is established. For the experimental results which are shown, the adaptively restored images have better quality than the nonadaptively restored ones based on visual observations and on an objective criterion of merit which accounts for the noise masking property of the visual system     

一种图像融合的新方法

在众多图像融合算法中,不存在完美的算法,每一种算法都具有优点和不足。本文针对这一问题提出一种新的图像融合方法,即图像二次融合,通过对两种算法的融合结果再进行一次简单的像素选择得到最终融合图像。图像二次融合可以综合算法的优点,弥补之间的不足,获得更好的融合效果。实验表明,该方法获得的融合结果优于单一算法。     

基于图切割的拉普拉斯金字塔图像融合算法

针对在图像拼接中普通的拉普拉斯金字塔融合算法容易丢失细节的缺点,以及由于运动物体造成的融合鬼影现象,本文提出了一种基于图切割的拉普拉斯金字塔融合算法。首先,引入图切割技术,寻找最优缝合线,确定一种自适应融合区,以消除运动物体造成的融合鬼影;其次,利用源图像完整细节对重构误差进行补偿,提出一种基于包含水平方向在内多个方向的加权融合方法,将源图像和拉普拉斯金字塔融合图像按照这种融合规则进行融合。实验结果表明,与经典拉普拉斯融合方法对比,在客观指标上,本文方法的图像均值平均提高了0.326,标准差(SD)平均提高了1.109,信息熵平均提高了0.041,图像清晰度平均提高了0.289;在主观效果上,本文方法无明显拼接痕迹和融合鬼影,较好保留了图像细节,提高了融合质量,全景图拼接更加真实,改善了整体的视觉效果。     

多输入融合对抗网络的水下图像增强

针对水下图像出现对比度低、颜色偏差和细节模糊等问题,提出了多输入融合对抗网络进行水下图像增强。该方法主要特点是生成网络采用编码解码结构,通过卷积层滤除噪声,利用反卷积层恢复丢失的细节并逐像素进行细化图像。首先,对原始图像进行预处理,得到颜色校正和对比度增强两种类型图像。其次,利用生成网络学习两种增强图像与原始图像之间差异的置信度图。然后,为减少在生成网络学习过程中两种增强算法引入的伪影和细节模糊,添加了纹理提取单元对两种增强图像进行纹理特征提取,并将提取的纹理特征与对应的置信度图进行融合。最后,通过构建多个损失函数,反复训练对抗网络,得到增强的水下图像。实验结果表明,增强的水下图像色彩鲜明并且对比度提升,评价指标UCIQE均值为0.639 9,NIQE均值为3.727 3。相比于其他算法有显著优势,证明了该算法的良好效果。     

基于对偶树复小波域HMT模型的遥感图像融合

针对遥感图像融合领域的实际应用,提出一种基于对偶树复小波变换与隐马尔可夫树模型结合的图像融合新方法。该算法将分别具有高光谱和高空间分辨率优势的两幅图像进行复小波变换,再对分解后不同频率域的系数选择不同的融合规则处理。采用低频系数加权平均;高频系数先建模,再基于区域能量规则处理的方法,最后完成逆变换得到重构图像。将该算法与其他几种图像融合方法进行比较,实验表明,该算法能够取得较为理想的效果。     

基于形态学增强和图像融合的板带钢缺陷检测

为了检测噪声和光照不均并存的多种类型的板带钢表面缺陷,提出了基于数学形态学增强和图像融合的缺陷检测算法。本文首先分别对图像作多结构形态学熵图像增强和多结构形态学边缘增强,其次对增强后的图像采用加权融合,并通过图像背景熵和增强图像的像素均值比确定权系数,最后对融合图像进行二值化处理以便于后续的缺陷识别及分类。 实验表明,本文算法不仅能准确检测出含有光照不均和大量噪声的板带钢图像中的表面缺陷,而且对于其他类型的板带钢缺陷图像也能获得较好的效果。除此之外,该算法具有较强的抗噪性和较高的稳定性。     

一种新的基于图像增强的融合算法

针对一般融合算法在图像预处理上存在的不足,将图像融合引入图像预处理中,使待融合图像不仅得到增强,且不损失其他信息,为下一步图像融合奠定良好的基础,在此基础上对图像进行融合,其标准差、平均梯度、熵等图像评价指标都优于直接对图像进行融合,达到预期效果.     

A class of robust entropic functionals for image restoration

This paper considers the concept of robust estimation in regularized image restoration. Robust functionals are employed for the representation of both the noise and the signal statistics. Such functionals allow the efficient suppression of a wide variety of noise processes and permit the reconstruction of sharper edges than their quadratic counterparts. A new class of robust entropic functionals is introduced, which operates only on the high-frequency content of the signal and reflects sharp deviations in the signal distribution. This class of functionals can also incorporate prior structural information regarding the original image, in a way similar to the maximum information principle. The convergence properties of robust iterative algorithms are studied for continuously and noncontinuously differentiable functionals. The definition of the robust approach is completed by introducing a method for the optimal selection of the regularization parameter. This method utilizes the structure of robust estimators that lack analytic specification. The properties of robust algorithms are demonstrated through restoration examples in different noise environments.     

基于自适应脉冲耦合神经网络图像融合新算法

针对传统的基于脉冲耦合神经网络(PCNN)的融合算法中每个神经元链接强度取同一常数的不足,提出了一种基于自适应PCNN图像融合新算法。作为显著性特征,使用像素的拉普拉斯能量(EOL,energy of Lapla-cian)和标准差(SD,standard deviation)分别作为PCNN对应神经元的链接强度值。实验结果表明,本文方法融合结果优于Laplacian方法、小波方法和传统的PCNN方法。     

Regularized adaptive high-resolution image reconstruction considering inaccurate subpixel registration

We propose a high-resolution image reconstruction algorithm considering inaccurate subpixel registration. A regularized iterative reconstruction algorithm is adopted to overcome the ill-posedness problem resulting from inaccurate subpixel registration. In particular, we use multichannel image reconstruction algorithms suitable for applications with multiframe environments. Since the registration error in each low-resolution image has a different pattern, the regularization parameters are determined adaptively for each channel. We propose two methods for estimating the regularization parameter automatically. The proposed algorithms are robust against registration error noise, and they do not require any prior information about the original image or the registration error process. Information needed to determine the regularization parameter and to reconstruct the image is updated at each iteration step based on the available partially reconstructed image. Experimental results indicate that the proposed algorithms outperform conventional approaches in terms of both objective measurements and visual evaluation.     

Spatially adaptive wavelet-based multiscale image restoration

In this paper, we present a new spatially adaptive approach to the restoration of noisy blurred images, which is particularly effective at producing sharp deconvolution while suppressing the noise in the flat regions of an image. This is accomplished through a multiscale Kalman smoothing filter applied to a prefiltered observed image in the discrete, separable, 2-D wavelet domain. The prefiltering step involves constrained least-squares filtering based on optimal choices for the regularization parameter. This leads to a reduction in the support of the required state vectors of the multiscale restoration filter in the wavelet domain and improvement in the computational efficiency of the multiscale filter. The proposed method has the benefit that the majority of the regularization, or noise suppression, of the restoration is accomplished by the efficient multiscale filtering of wavelet detail coefficients ordered on quadtrees. Not only does this lead to potential parallel implementation schemes, but it permits adaptivity to the local edge information in the image. In particular, this method changes filter parameters depending on scale, local signal-to-noise ratio (SNR), and orientation. Because the wavelet detail coefficients are a manifestation of the multiscale edge information in an image, this algorithm may be viewed as an "edge-adaptive" multiscale restoration approach.     

Memristor-based multi-channel pulse coupledneural network for image fusion

Image fusion is widely used in computer vision and image analysis. Considering that the traditional image fusionalgorithm has a certain limitation in multi-channel image fusion, a memristor-based multi-channel pulse coupledneural network (M-MPCNN) for image fusion is proposed. Based on a dual-channel pulse coupled neural network(D-PCNN), a novel multi-channel pulse coupled neural network (M-PCNN) is firstly constructed in this paper.Then the exponential growth dynamic threshold model is used to improve the pulse generation of pulse coupledneural network, which can not only avoid multiple ignitions effectively, but can also improve operational efficiencyand reduce complexity. At the same time, synchronous capture can also enhance image edge, which is moreconducive to image fusion. Finally, the threshold and synaptic characteristics of pulse coupled neural networks(PCNNs) can be well realized by using a memristor-based pulse generator. Experimental results show that theproposed algorithm can fuse multi-source images more effectively than existing state-of-the-art fusion algorithms.     

基于TwIST-TV 约束的图像去模糊方法

传统的基于频域和小波域的去模糊算法所得的复原图像总是存在比较明显的边缘振铃及模糊效应,而较为有效的空域迭代优化去模糊算法速度通常比较慢。为了解决上述问题,提出了基于二步迭代阈值收缩（TwIST）与总变分（TV）约束相结合的图像去模糊算法（TwIST-TV）。首先在去模糊目标函数中加入对图像的TV 正则化约束,其次在对图像小波系数的每次二步迭代之前,加入对图像的TV 优化去噪约束,最后迭代获取去模糊图像。实验结果表明:相对于基于频域和小波域的模糊图像恢复算法,TwIST-TV 能有效抑制边缘模糊和振铃效应,复原图像的信噪比（SNR）、峰值信噪比（PSNR）高出1~7 dB,平均结构相似度指标（MSSIM）可高出0.05,相对于空域解卷积算法在保证求解精度相当的情况下具备6 倍以上的速度优势。