基于并行压缩感知理论的红外夜天文图像超分辨率成像方法研究

项目名称： 基于并行压缩感知理论的红外夜天文图像超分辨率成像方法研究

项目编号： No.11503010

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 天文学、地球科学

项目作者： 隋修宝

作者单位： 南京理工大学

项目金额： 22万元

中文摘要： 超分辨率红外夜天文观测在揭示冷暗物质、探索隐藏宇宙、追溯宇宙早期生命等方面具有广泛的应用。然而红外夜天文观测的独有特点：大气扰动导致图像序列之间产生位移且位移量难以估计，红外镜头移动不便和缺少高分辨率图像作为先验知识，使得以位移估计、先验学习和基于图像序列位移不变性的压缩感知为代表的超分辨率成像方法难以满足天文观测的实际应用。为此，本项目提出了并行压缩感知的概念，并探索以此为理论基础的超分辨率成像方法。思路为：通过研究特殊的测量矩阵，实现测量数据的并行获取，从而单次测量便获得超分辨率重建所需的所有数据；结合测量值的非均匀性校正和强噪声条件下的恢复算法，实现红外夜天文图像的超分辨率重建。该方法能有效提高测量数据的获取效率，避免传统压缩感知方法对图像序列位移不变性的要求，实现单帧超分辨率重建，而且无需位移估算和先验学习，可望应用于红外天文领域的超分辨率成像，具有重要的学术价值和广阔的应用前景。

中文关键词： 红外夜天文成像；超分辨率重建；并行压缩感知；非均匀性校正

英文摘要： Super-resolution infrared night astronomical observations have revealed a wide application in discovering hidden galaxies, exploring the unknown universe, tracing the early universe life, and other aspects. However, the unique characteristic of infrared night astronomical observation is that atmospheric disturbance generates unestimated displacement between the image sequences, infrared lens are moved uneasily, and the observation lacks of high-resolution image as the prior knowledge, which makes the traditional super-resolution imaging method hard to satisfy the practical application of astronomical observation. The traditional super-resolution imaging method mainly include the methods based on displacement estimation, prior learning, and compressed sensing which assumes the displacement is invariant between image sequences. Therefore, we present the concept of parallel compressed sensing, and discover a new super-resolution method based on this theory. The idea is that, by discovering the spatial measurement matrix to achieve the image data in parallel, we can get the whole reconstruction data in one single measurement. Then combining with the nonuniformity correction of measurement data and the recovery algorithms in high noise environment, we can reconstruct the super-resolution infrared night astronomical image. This method can effectively raise the data acquisition efficiency; avoid the requirement of invariant displacement between image sequences in traditional compressed sensing methods, and achieve single-frame super-resolution reconstruction without displacement estimation and prior learning. It is expected to be applied in the field of infrared astronomy super-resolution imaging, and has important academic value and broad application prospects.

英文关键词： infrared night astronomical imaging;super-resolution reconstruction;parallel compressed sensing;nonuniformity correction