This paper presented a novel method on designing redundant dictionary from known orthogonal functions. Usual way of discretization of continuous functions is uniform sampling. Our experiments show that dividing the fu...This paper presented a novel method on designing redundant dictionary from known orthogonal functions. Usual way of discretization of continuous functions is uniform sampling. Our experiments show that dividing the function definition interval with non-uniform measure makes the redundant dictionary sparser and it is suitable for image denoising via sparse and redundant dictionary. In this case the problem is to find an appropriate measure in order to make each atom of dictionary. It has shown that in sparse approximation context, incoherent dictionary is suitable for sparse approximation method. According to this fact we define some optimization problems to find the best parameter of distribution measure (in our study normal distribution). For better convergence to optimum point we used Genetic Algorithm (GA) with enough diversity on initial population. We show the effect of this type of dictionary design on exact sparse recovery support. Our results also show the advantage of this design method on image denoising task.展开更多
Image denoising is the basic problem of image processing. Quaternion wavelet transform is a new kind of multiresolution analysis tools. Image via quaternion wavelet transform, wavelet coefficients both in intrascale a...Image denoising is the basic problem of image processing. Quaternion wavelet transform is a new kind of multiresolution analysis tools. Image via quaternion wavelet transform, wavelet coefficients both in intrascale and in interscale have certain correla- tions. First, according to the correlation of quaternion wavelet coefficients in interscale, non-Ganssian distribution model is used to model its correlations, and the coefficients are divided into important and unimportance coefficients. Then we use the non-Gaussian distribution model to model the important coefficients and its adjacent coefficients, and utilize the MAP method estimate original image wavelet coefficients from noisy coefficients, so as to achieve the purpose of denoising. Experimental results show that our al- gorithm outperforms the other classical algorithms in peak signal-to-noise ratio and visual quality.展开更多
Shearlet变换作为后小波时代的一个重要的多尺度几何分析工具具有良好的各向异性和方向捕捉性,同时它也可以对诸如图像等多维信号进行一种近最优的稀疏表示.非下采样Shearlet变换(NSST)在保持Shearlet变换特性的同时还具有平移不变特性...Shearlet变换作为后小波时代的一个重要的多尺度几何分析工具具有良好的各向异性和方向捕捉性,同时它也可以对诸如图像等多维信号进行一种近最优的稀疏表示.非下采样Shearlet变换(NSST)在保持Shearlet变换特性的同时还具有平移不变特性,这在具有丰富纹理和细节信息的图像处理中发挥着重要作用.该文首先对图像NSST方向子带内系数的概率密度分布进行分析,获得系数的稀疏统计特性和Cauchy分布拟合子带内系数的有效性;其次对NSST方向子带间系数的联合概率分布进行分析,获得方向子带系数间所具有的持续和传递特性,确定了一种NSST子带间树形架构的系数对应关系,进而提出一种NSST域隐马尔可夫模树模型(C-NSSTHMT),该模型通过Cauchy分布来拟合NSST系数,更好地揭示图像NSST变换后相同尺度子带内和不同尺度子带间系数的相关性.进一步提出一种基于所提出C-NSST-HMT模型的图像去噪算法,该算法对于含噪声方差为30和40的噪声图像,其去噪后的PSNR(Peak Signal to Noise Ratio)较NSCT-HMT方法分别提高了1.995dB和1.193dB.特别对纹理和细节丰富的图像,该算法在去噪的同时,有效地保留了图像的几何信息.展开更多
针对磁共振(magnetic resonance,MR)幅度图像中带有不易去除的与信号相关的莱斯(Rician)噪声问题,利用其复数图像中的实部与虚部所含噪声为不相关的加性高斯白噪声这一特性,代替对幅度图像直接去噪,提出将原始对偶字典学习(predual dict...针对磁共振(magnetic resonance,MR)幅度图像中带有不易去除的与信号相关的莱斯(Rician)噪声问题,利用其复数图像中的实部与虚部所含噪声为不相关的加性高斯白噪声这一特性,代替对幅度图像直接去噪,提出将原始对偶字典学习(predual dictionary learning,PDL)算法用于对MR复数图像的实部与虚部分别进行去噪,然后组合得到幅度图像的方法.经仿真实验和在HT-MRSI50-50(50 mm)1.2 T小动物核磁共振系统中的实际应用,证明所提方法较直接对幅度图像去噪取得更好的效果,在有效去除MR图像噪声的同时能较好地保持图像中的细节.与经典的字典学习算法核奇异值分解(kernel singular value decomposition,K-SVD)相比,PDL算法去噪效果优于K-SVD算法,而运算速度提高约5倍.与经典的基于非局部相似块的三维块匹配滤波(block-matching and 3D filtering,BM3D)算法相比,在噪声水平较低时PDL算法略优于BM3D算法,噪声水平较高时BM3D算法略优于PDL算法,两者总体比较接近.展开更多
文摘This paper presented a novel method on designing redundant dictionary from known orthogonal functions. Usual way of discretization of continuous functions is uniform sampling. Our experiments show that dividing the function definition interval with non-uniform measure makes the redundant dictionary sparser and it is suitable for image denoising via sparse and redundant dictionary. In this case the problem is to find an appropriate measure in order to make each atom of dictionary. It has shown that in sparse approximation context, incoherent dictionary is suitable for sparse approximation method. According to this fact we define some optimization problems to find the best parameter of distribution measure (in our study normal distribution). For better convergence to optimum point we used Genetic Algorithm (GA) with enough diversity on initial population. We show the effect of this type of dictionary design on exact sparse recovery support. Our results also show the advantage of this design method on image denoising task.
基金Supported by Natural Science Foundation of Anhui (No.11040606M06)
文摘Image denoising is the basic problem of image processing. Quaternion wavelet transform is a new kind of multiresolution analysis tools. Image via quaternion wavelet transform, wavelet coefficients both in intrascale and in interscale have certain correla- tions. First, according to the correlation of quaternion wavelet coefficients in interscale, non-Ganssian distribution model is used to model its correlations, and the coefficients are divided into important and unimportance coefficients. Then we use the non-Gaussian distribution model to model the important coefficients and its adjacent coefficients, and utilize the MAP method estimate original image wavelet coefficients from noisy coefficients, so as to achieve the purpose of denoising. Experimental results show that our al- gorithm outperforms the other classical algorithms in peak signal-to-noise ratio and visual quality.
文摘磁共振成像(Magnetic Resonance Imaging,MRI)已经成为一种常见的影像检查方式,MRI的去噪算法影响着MRI的成像效果。基于深度学习的MRI去噪算法需要一定量的数据,绝大部分基于非深度学习的MRI去噪算法都是将MRI数据转化为实数之后进行去噪的,针对复数MRI中的复数数据类型的算法也存在着失真的问题。因此,提出一种通过单张MRI脑图像的原始数据进行噪点剔除的算法,以此更好得去除图像噪声。该算法从MRI的原始数据出发,利用了MRI噪声分布性质和MRI脑图像的特点,以判断MRI图像中噪声明显的点,从而剔除MRI中特定的莱斯分布的噪声。并将所提出的算法结合了MRI去噪中常用的非局部平均算法(Non-Local Means denoising,NLM)与三维块匹配算法(Block-Matching and 3D filtering,BM3D),并和不使用该算法剔除噪点的NLM、BM3D进行了对比评估。对比结果表明,在噪声密度不同的多种情况下,该算法总能优化与之相结合的图像去噪算法,在不同的噪声情况下使峰值信噪比(Peak Signal-to-Noise Ratio,PSNR)与结构相似性(Structural Similarity,SSIM)提高了1%~9%。最后将该算法结合BM3D,对比了DnCNN、低秩聚类算法(Weighted Nuclear Norm Minimization,WNNM)、BM3D、NLM等用于MRI去噪的算法,在莱斯噪声较多时,该算法在PSNR上有更好的表现。
文摘Shearlet变换作为后小波时代的一个重要的多尺度几何分析工具具有良好的各向异性和方向捕捉性,同时它也可以对诸如图像等多维信号进行一种近最优的稀疏表示.非下采样Shearlet变换(NSST)在保持Shearlet变换特性的同时还具有平移不变特性,这在具有丰富纹理和细节信息的图像处理中发挥着重要作用.该文首先对图像NSST方向子带内系数的概率密度分布进行分析,获得系数的稀疏统计特性和Cauchy分布拟合子带内系数的有效性;其次对NSST方向子带间系数的联合概率分布进行分析,获得方向子带系数间所具有的持续和传递特性,确定了一种NSST子带间树形架构的系数对应关系,进而提出一种NSST域隐马尔可夫模树模型(C-NSSTHMT),该模型通过Cauchy分布来拟合NSST系数,更好地揭示图像NSST变换后相同尺度子带内和不同尺度子带间系数的相关性.进一步提出一种基于所提出C-NSST-HMT模型的图像去噪算法,该算法对于含噪声方差为30和40的噪声图像,其去噪后的PSNR(Peak Signal to Noise Ratio)较NSCT-HMT方法分别提高了1.995dB和1.193dB.特别对纹理和细节丰富的图像,该算法在去噪的同时,有效地保留了图像的几何信息.
文摘针对磁共振(magnetic resonance,MR)幅度图像中带有不易去除的与信号相关的莱斯(Rician)噪声问题,利用其复数图像中的实部与虚部所含噪声为不相关的加性高斯白噪声这一特性,代替对幅度图像直接去噪,提出将原始对偶字典学习(predual dictionary learning,PDL)算法用于对MR复数图像的实部与虚部分别进行去噪,然后组合得到幅度图像的方法.经仿真实验和在HT-MRSI50-50(50 mm)1.2 T小动物核磁共振系统中的实际应用,证明所提方法较直接对幅度图像去噪取得更好的效果,在有效去除MR图像噪声的同时能较好地保持图像中的细节.与经典的字典学习算法核奇异值分解(kernel singular value decomposition,K-SVD)相比,PDL算法去噪效果优于K-SVD算法,而运算速度提高约5倍.与经典的基于非局部相似块的三维块匹配滤波(block-matching and 3D filtering,BM3D)算法相比,在噪声水平较低时PDL算法略优于BM3D算法,噪声水平较高时BM3D算法略优于PDL算法,两者总体比较接近.
