统计与结构先验联合利用的压缩感知图像重构
发布时间：2018-08-27 11:19
【摘要】：在信息技术高速发展的今天,图像作为最直观的信息载体之一,已成为数据传输的主流形式。随着人们对图像质量的要求不断提高,对数据的需求量越来越大,传统图像压缩与传输技术已难以满足日益增长的数据带宽需求。压缩感知(Compressed Sensing,CS)理论应运而生。该理论突破传统信号算法中采样速率需遵循奈奎斯特采样定理(Nyquist Sampling Theorem)的约束,根据信号的稀疏性或可压缩性,对信号进行低速压缩采样,采样频率远低于奈奎斯特采样定律,并运用重构算法准确(针对稀疏信号)或近似(针对可压缩信号)准确地重构出原始信号。基于压缩感知理论的图像压缩与重构算法能够有效地节省编码端采样和压缩的资源成本,从而在数据量大且冗余度高的图像信号压缩与传输领域有着广阔的应用前景,成为该领域学者的研究热点。图像信号所具备的变换域稀疏性(sparsity)或可压缩性(compressibility)为压缩感知理论的应用提供了前提保证。传统CS图像重构算法仅考虑了图像信号在小波域等变换域上的稀疏特性或可压缩特性,但并未充分考虑对其统计特性和结构特性的充分利用。图像的小波稀疏表示形式除稀疏性外,还具有较强的类聚性(cluster),这表现在图像信号经小波变换后,稀疏表示系数层间呈现的树状结构关系及层内表现出的统计相依分布。本文针对图像小波稀疏表示系数的特性,从统计先验角度和结构先验角度分别对图像进行模型分析及研究,并将结构先验模型与统计先验模型分别融入经典重构算法中,取得了较好的重构效果。为进一步提高图像压缩感知重构算法的重构质量与效率,本文创新性地提出统计与结构先验联合利用的CS图像重构算法,对图像信号的稀疏表示形式进行层内层间建模,利用多重先验信息对经典重构算法进行优化:针对图像小波表示系数的层内层间关系,利用高斯尺度混合模型对系数局部建模,并利用系数层间树结构模型对其进行结构约束,应用迭代阈值算法求解稀疏表示系数估计值,最终利用少量采样值实现高质高效的图像重构。本文对所提算法以及经典CS图像重构算法进行仿真比较,测试结果表明,联合利用统计与结构先验的CS图像重构算法在图像的重构性能上有明显优化。对于重构精度,峰值信噪比较单一模型下的重构算法最高提升4dB左右,重构速度也有较大提升,是集高效性与实用性为一体的CS图像重构算法。
[Abstract]:With the rapid development of information technology, image, as one of the most intuitive information carriers, has become the mainstream form of data transmission. With the increasing demand for image quality and the increasing demand for data, the traditional image compression and transmission technology is difficult to meet the increasing demand of data bandwidth. The theory of compressed perception (Compressed Sensing,CS) came into being. This theory breaks through the constraint of Nyquist sampling theorem (Nyquist Sampling Theorem) in the traditional signal algorithm. According to the sparsity or compressibility of the signal, the signal is compressed at low speed and the sampling frequency is much lower than that of Nyquist sampling law. The original signal is reconstructed accurately by using reconstruction algorithm (for sparse signal) or approximate (for compressible signal). The algorithm of image compression and reconstruction based on the theory of compression perception can save the cost of sampling and compression in the coding end effectively, so it has a broad application prospect in the field of image signal compression and transmission, which has a large amount of data and high redundancy. It has become the research hotspot of scholars in this field. The transform domain sparse (sparsity) or compressible (compressibility) of image signal provides a prerequisite for the application of compression sensing theory. The traditional CS image reconstruction algorithm only considers the sparse or compressible characteristics of the image signal in the domain of wavelet transform, but does not fully consider the full use of its statistical and structural characteristics. In addition to sparsity, the wavelet sparse representation of images also has a strong clustering (cluster),. After wavelet transform, the sparse representation shows the tree structure relationship among the coefficient layers and the statistical dependence distribution in the layers. In this paper, according to the characteristics of sparse representation coefficients of image wavelet, the image model is analyzed and studied from the perspective of statistical priori and structural priori, and the structural priori model and statistical priori model are incorporated into the classical reconstruction algorithm, respectively. Good reconstruction effect has been achieved. In order to improve the reconstruction quality and efficiency of the image compression perceptual reconstruction algorithm, a novel CS image reconstruction algorithm based on statistical and structural priori is proposed in this paper, in which the sparse representation of image signals is modeled between layers. The classical reconstruction algorithm is optimized by using multiple prior information. According to the interlayer relationship of image wavelet representation coefficient, Gao Si scale mixed model is used to model the coefficient locally, and the coefficient interlayer tree structure model is used to constrain the coefficient. An iterative threshold algorithm is used to solve the estimation of sparse representation coefficients, and a small number of sampling values are used to achieve high quality and efficient image reconstruction. In this paper, the proposed algorithm and the classical CS image reconstruction algorithm are simulated and compared. The test results show that the combined use of statistical and structural priori CS image reconstruction algorithm has obvious optimization in image reconstruction performance. For the reconstruction accuracy, the peak signal-to-noise ratio (PSNR) algorithm can improve the 4dB and the reconstruction speed greatly. It is a CS image reconstruction algorithm with high efficiency and practicability.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41
本文编号：2207120
[Abstract]:With the rapid development of information technology, image, as one of the most intuitive information carriers, has become the mainstream form of data transmission. With the increasing demand for image quality and the increasing demand for data, the traditional image compression and transmission technology is difficult to meet the increasing demand of data bandwidth. The theory of compressed perception (Compressed Sensing,CS) came into being. This theory breaks through the constraint of Nyquist sampling theorem (Nyquist Sampling Theorem) in the traditional signal algorithm. According to the sparsity or compressibility of the signal, the signal is compressed at low speed and the sampling frequency is much lower than that of Nyquist sampling law. The original signal is reconstructed accurately by using reconstruction algorithm (for sparse signal) or approximate (for compressible signal). The algorithm of image compression and reconstruction based on the theory of compression perception can save the cost of sampling and compression in the coding end effectively, so it has a broad application prospect in the field of image signal compression and transmission, which has a large amount of data and high redundancy. It has become the research hotspot of scholars in this field. The transform domain sparse (sparsity) or compressible (compressibility) of image signal provides a prerequisite for the application of compression sensing theory. The traditional CS image reconstruction algorithm only considers the sparse or compressible characteristics of the image signal in the domain of wavelet transform, but does not fully consider the full use of its statistical and structural characteristics. In addition to sparsity, the wavelet sparse representation of images also has a strong clustering (cluster),. After wavelet transform, the sparse representation shows the tree structure relationship among the coefficient layers and the statistical dependence distribution in the layers. In this paper, according to the characteristics of sparse representation coefficients of image wavelet, the image model is analyzed and studied from the perspective of statistical priori and structural priori, and the structural priori model and statistical priori model are incorporated into the classical reconstruction algorithm, respectively. Good reconstruction effect has been achieved. In order to improve the reconstruction quality and efficiency of the image compression perceptual reconstruction algorithm, a novel CS image reconstruction algorithm based on statistical and structural priori is proposed in this paper, in which the sparse representation of image signals is modeled between layers. The classical reconstruction algorithm is optimized by using multiple prior information. According to the interlayer relationship of image wavelet representation coefficient, Gao Si scale mixed model is used to model the coefficient locally, and the coefficient interlayer tree structure model is used to constrain the coefficient. An iterative threshold algorithm is used to solve the estimation of sparse representation coefficients, and a small number of sampling values are used to achieve high quality and efficient image reconstruction. In this paper, the proposed algorithm and the classical CS image reconstruction algorithm are simulated and compared. The test results show that the combined use of statistical and structural priori CS image reconstruction algorithm has obvious optimization in image reconstruction performance. For the reconstruction accuracy, the peak signal-to-noise ratio (PSNR) algorithm can improve the 4dB and the reconstruction speed greatly. It is a CS image reconstruction algorithm with high efficiency and practicability.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41
【参考文献】
相关期刊论文 前3条
1 何宜宝;毕笃彦;;基于广义拉普拉斯分布的图像压缩感知重构[J];中南大学学报(自然科学版);2013年08期
2 练秋生;肖莹;;基于小波树结构和迭代收缩的图像压缩感知算法研究[J];电子与信息学报;2011年04期
3 练秋生;王艳;;基于双树小波通用隐马尔可夫树模型的图像压缩感知[J];电子与信息学报;2010年10期
，本文编号：2207120
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2207120.html
