变分贝叶斯超分辨率重建算法研究-光学工程专业论文.docxVIP

摘要 摘要 摘要 超分辨率重建技术是指利用图像处理技术，从拍摄到的一组低分辨率图像中重 建出一幅高分辨率图像，增强图像中的高频分量并消除由低分辨率成像系统产生 的图像退化，突破硬件系统对图像分辨率的限制，在图像处理领域得到了广泛的 应用。 本文首先介绍了超分辨率重建算法的数学模型、主要算法以及目前面临的一些 挑战。然后针对大多数超分辨率重建算法对噪声和运动估计误差稳定性较差的缺 陷，详细阐述了变分贝叶斯超分辨率重建算法。变分贝叶斯算法把未知的高分辨 率图像、运动参数和未知的模型参数在一个分层贝叶斯框架下建模。该框架对未 知变量进行概率分布估计而不是点估计，在估计过程中引入了不确定性，阻止不 同变量间估计误差的扩大。把运动参数估计和算法参数的自适应计算加入到重建 模型中，提高了算法对噪声的稳定性，并使算法不需要进行参数调整。最后通过 实验结果的对比可以看到，变分贝叶斯超分辨率重建算法取得了较好的重建结果。 关键字：超分辨率重建 贝叶斯方法 参数估计 变分分析 Ab Abstract Abstract Super resolution reconstruction are techniques that construct high resolution images from a set of observed low resolution images through image processing, thereby increasing the high frequency components, removing the degradations caused by the low resolution imaging system and breaking through the resolution limitations of the hardware system. Now super resolution has already obtained a wide range of applications in the field of image processing. At first, this paper describes the mathematical models of super resolution algorithm, common algorithms and some challenges currently facing. Then, for solving the problem that most super resolution algorithms is vulnerable to noise and estimation errors of motion parameters, variational Bayesian super resolution algorithm is elaborated. This algorithm models the unkonwn high resolution image, motion parameters and unkonwn model parameters in a hierarchical Bayesian framework. This framework allows for distribution estimation of unkonwns, instead of point estimation, which incorporates uncertainty of estimates and prevent the propagation of errors between the estimates of the various unknowns. Through the incorporation of motion estimation and adaptive estimation of the algorithm parameters, the algorithm is very robust to noise and errors in motion estimation, and does not require parameter tuning. Finally, the contrast experiments demonstrates that variational Bayesian algorithm can provide better super resolution results. Keywords: super resolution Bayesian methods