【作者】 李林静；

【导师】 冯勇；

【作者基本信息】 哈尔滨工业大学 ， 控制科学与工程， 2007， 硕士

【摘要】 图像的超分辨率重构是指利用同一景物的一幅或多幅低分辨率图像来估计该景物的较高分辨率图像。在不提高图像获取设备物理分辨率的前提下,该方法能通过软件算法提高图像的分辨率。在遥感图像处理、军事侦查、医学成像、机器视觉、公共安全和多媒体电子消费等领域有着广泛的应用。本文研究图像超分辨率重构的快速算法。首先,分析了图像退化的向量式模型,通过矩阵的张量积和矩阵空间跟向量空间的同构映射,将向量式模型转换为等价的矩阵式模型,与向量式模型相比,本文模型节省了计算所需的存储量,且节省量正比于降采样因子和原始低分辨率图像大小的乘积。然后,在矩阵式图像退化模型的基础上,利用最速下降法和共轭梯度法来求取图像超分辨率重构问题的最优解。接着,提出了一种快速迭代式投影重构算法,通过在矩阵空间中定义一个求解重构问题的投影算子,并利用高通滤波器来获取并增强图像的高频信息,以提高图像的分辨率。最后,设计了图像超分辨率重构软件,用于实现上述重构算法。计算机仿真结果证明了所提算法的正确性,且计算速度与现有算法相比有较大的提高,可以直接应用于大尺寸图像的重构。综合考虑计算时间和重构质量,本文所研究的迭代投影算法具有较好的应用前景。更多还原

【Abstract】 Super-resolution image reconstruction is a technique which can obtain a high resolution image from one or multiple warped, blurred and noised low resolution image(s). It can increase the spatial resolution of image by software without updating hardware system. So it has a wide application in remote sensing, military detection, medical imaging, machine vision, public security, multimedia E-commerce and etc.This dissertation is concentrate on fast image SR algorithms. First, the matrix based image degrading model is obtained with the aid of matrix tensor product and the isomorphision that between matrix and vector space. Matrix model need less memory and analysis shows the saved memory is proportional to the product of subsampling factor and the dimension of low resolution image. Second, two fast SR algorithms called Matrix-Gradient-Descent and Matrix-Conjugate-Gradient which using the matrix as their variable are proposaled under the regular theory and matrix degrading model. Third, a projective operator is defined on the matrix space which can get the high resolution image from any other estimation directly. Combine with a Laplace high pass filter, high-frequency components can be got and boosted, this can be used to increase the spatial resolution of a image, based on this a fast iterative projection SR reconstruction algorithm is proposaled. At the end, SR software is designed to realize the proposed algorithms.The validity of proposaled algorithms is proved by computer simulation. And, low resolution image with large dimensions can be reconstuctioned directly by these algorithms for their faster speed. The iterative projection SR algorithm has the highest cost-performance when a trade off exists between the computing time and reconstruction quality.更多还原