红外成像超分辨率图像重建算法研究

	红外成像超分辨率图像重建算法研究
其他题名	Research on infrared image super-resolution reconstruction
	何阳
学位类型	硕士
导师	黄玮
	2015-11
学位授予单位	中国科学院大学
学位专业	光学
关键词	红外图像字典学习 Omp算法迭代反投影凸集投影
摘要	红外成像系统的分辨率受限于红外探测器的限制，，红外成像光学系统难以获得高分辨率的红外图像。由于红外探测器硬件条件的限制，难以从硬件方面提高图像分辨率，可以利用软件绕过硬件的技术瓶颈来提高红外图像的分辨率。分别对单帧低分辨率红外图像输入和序列低分辨率红外图像输入的两类图像超分辨率重建算法进行了研究。其一利用基于字典学习的方法实现单帧低分辨率图像输入的超分辨率图像重建，其二利用凸集投影算法(Projection Onto Convex Sets, POCS)实现序列图像输入的超分辨率图像重建。对上述两种算法，采用峰值信噪比、结构自相似度以及视觉直观观察对重建图像结果进行评价，说明了上述两种算法能够有效地实现红外图像超分辨率重建。基于字典学习的方法是通过对一组样本图像进行训练分析，得到高低分辨率图像之间的联系规律，然后根据这一规律对低分辨率图像进行超分辨率重建。在字典学习算法的高低分辨率字典构建上，由于其数据量十分巨大，采用主要元素分析(Principal Component Analysis ,PCA)构造子矩阵方法以及K-SVD算法来降低计算复杂度。在对图像进行稀疏表示的过程中，利用稀疏阈值的正交匹配追踪(Orthogonal Matching Pursuit algorithm, OMP)算法来降低计算复杂度。并且根据重建图像的残差信息，通过改进的迭代反投影法进一步改善重建图像质量。对序列低分辨率红外图像输入采用基于字典学习去噪的凸集投影算法进行超分辨率重建。对传统的凸集投影算法改进和优化，在图像参考帧修正过程中，通过加入边缘检测算子，可以将图像划分成平滑区域与非平滑区域，采用不同的修正点扩散函数(Point Spread Function, PSF)权重系数对图像进行修正。考虑到凸集投影算法难以对噪声处理，引入K-SVD字典学习方法进行去噪处理，增强凸集投影算法对噪声处理不足的情况，以达到提高高分辨率红外图像重建质量的目的。
其他摘要	The resolution of infrared imaging system is limited by the infrared detector. The imaging system of infrared imaging system has not produced a fundamental change. Because of the limitation of the hardware, it is difficult to improve the image resolution on the hardware, improving the resolution of the infrared image by using the technology of software is a good choice. This paper mainly from the single frame low resolution infrared image input and a sequence of infrared images of low resolution input two aspects to proceed, using single frame of the input image super resolution image reconstruction based on dictionary learning method, projection onto convex sets (POCS) algorithm is used to input image sequence super resolution image reconstruction. The two algorithms are evaluated by the peak signal to noise ratio, the structure self-similarity and the visual observation. The above two algorithms are described. The method based on dictionary learning is to analyze the image of a set of training samples and get the relation between the high and low resolution images. Then, the low resolution image is reconstructed by using the rule. In the dictionary learning algorithm of the high and low resolution dictionary construction, because of its huge data volume, the use of principal component analysis (PCA) for construction the sub matrix method and K-SVD algorithm to reduce the computational complexity. In the process of sparse representation of the image, the OMP algorithm is used to reduce the computational complexity. And according to the residual information of the reconstructed image, the reconstructed image quality is improved by using the modified iterative back projection method. Super resolution reconstruction of the set of low resolution infrared images input using POCS algorithm. The traditional POCS algorithm is improved and optimized in the process of image reference frame correction. By using edge detection operator, the image is divided into smooth region and non-smooth region. The image is modified by different modified PSF weight coefficients. Taking into account that the POCS algorithm is difficult to deal with the noise, the K-SVD dictionary learning method is introduced to improve the quality of high resolution infrared image reconstruction.
语种	中文
文献类型	学位论文
条目标识符	http://ir.ciomp.ac.cn/handle/181722/49236
专题	中科院长春光机所知识产出
推荐引用方式 GB/T 7714	何阳. 红外成像超分辨率图像重建算法研究[D]. 中国科学院大学,2015.

条目包含的文件
文件名称/大小	文献类型	版本类型	开放类型	使用许可
何阳.pdf（2350KB）	学位论文		开放获取	CC BY-NC-SA	请求全文