自适应红外图像非均匀性校正方法研究
文献类型：学位论文
| 作者 | 于世孔
|
| 学位类别 | 硕士 |
| 答辩日期 | 2015-05-26 |
| 授予单位 | 中国科学院沈阳自动化研究所 |
| 授予地点 | 中国科学院沈阳自动化研究所 |
| 导师 | 向伟 |
| 关键词 | 红外焦平面阵列 非均匀性校正 神经网络 鬼影 |
| 其他题名 | Research of Adaptive Nonuniformity Correction for Infrared Images |
| 学位专业 | 控制理论与控制工程 |
| 中文摘要 | 红外焦平面阵列（IRFPA）是目前最先进也是主流的红外探测器，广泛应用于各种军用和民用红外成像系统。但由于受材料提纯和器件制造工艺水平等因素所限，红外焦平面阵列普遍存在较强的非均匀性，严重影响红外成像质量，必须对输出图像进行非均匀性校正。因此，红外图像非均匀性校正技术已成为红外成像的关键技术之一。 基于标定的红外图像非均匀性校正方法原理简单、易实时实现，在工程上应用较多。但是，由于红外探测器在长时间使用或经过长期存放后，探测器特性会发生变化，导致成像质量下降，需重新进行标定，这在很大程度上制约了红外成像技术的应用发展。基于场景的自适应非均匀性校正方法不需要标定，成像系统可长时间连续工作，已成为研究的热点。 本文首先介绍了课题研究背景、红外图像非均匀性校正方法的研究现状以及红外图像非均匀性产生的机理，并对两点标定法、时域高通滤波法、神经网络法、恒定统计法、代数法和中值红外直方图均衡法等非均匀性校正算法进行了详细的分析和深入的研究。然后重点研究了神经网络法及鬼影问题产生机理，并在此基础上提出了两种改进的基于神经网络的自适应非均匀性校正算法，即基于张量约束双边滤波器的自适应非均匀性校正算法和基于高斯低通—三边联合滤波器的自适应非均匀性校正算法。经仿真和实际红外图像序列实验验证表明，所提算法均获得了良好的抑制鬼影效果。 同时，针对工程应用对实时非均匀性校正的需求，本文对自适应校正方法的硬件实现方案进行了探索。以FPGA为平台，提出了双边滤波器等算法核心模块的实现方法，在此基础上提出了一种非均匀性校正逻辑设计方案。仿真实验表明该设计能够满足实时图像处理要求，为自适应非均匀性校正方法的工程应用打下了基础。 |
| 索取号 | TP391.41/Y74/2015 |
| 英文摘要 | Infrared Focal Plane Array (IRFPA) is the most advanced and mainstream infrared detectors, and it has been widely used in military and civil infrared imaging systems. Under the restriction of the level of material refinement and device manufacturing process, strong nonuniformity exists in IRFPA widely and degenerates the quality of infrared imaging heavily. So Nonuniformity Correction (NUC) must be imposed on infrared images. Because of it, NUC for infrared images has been one of the most key technologies in infrared imaging. Method of calibration based NUC for infrared images has simple principles and is easy for real-time processing, so it has been widely applied in engineering. But the characteristic of the infrared detectors will be changed when the infrared detectors work or deposit for a long time which degenerates the quality of imaging. So the IRFPA must be calibrated periodically which constrains the development of the technology of infrared imaging for application to great degree. Scene based adaptive NUC doesn’t need calibration which makes it possible for the infrared imaging work continuously. This method has been a research focus in NUC. First, the researching background of this paper and NUC algorithms for infrared images are introduced. Second, the mechanism of the nonuniformity is analyzed. On the basis of it, some NUC algorithms, such as Two Points NUC (TPNUC), Temporal High-pass Filtering NUC (THPF-NUC), Neural Network NUC (NN-NUC), Constant Statistics NUC (CS-NUC), Algebraic NUC and Midway Infrared Equalization NUC (MIRE), are analyzed and researched in detail. Then, the NN-NUC is emphasized and the mechanism of ghosting artifacts is researched. On the basis of it, two improved neural network based adaptive NUC algorithms are proposed. They are tensor constrained bilateral filter based adaptive NUC algorithm and Gaussian low-pass & trilateral filter based adaptive NUC algorithm. After processing by simulation and real image sequences, the proposed algorithms have a good effect of preventing ghosting artifacts. At the same time, to satisfy the need of real-time NUC in engineering application, this paper tries to make an exploration of hardware implementation of adaptive NUC algorithms. Based on the platform of FPGA, methods for implementation of key modules like bilateral filter are designed. On the basis of it, a designing scheme of NUC is proposed. It is proved by simulation experiment that this design can satisfy the requirement of real-time image processing and lays a basis for the engineering application of adaptive NUC algorithms. |
| 语种 | 中文 |
| 产权排序 | 1 |
| 页码 | 75页 |
| 源URL | [http://ir.sia.ac.cn/handle/173321/16781]  |
| 专题 | 沈阳自动化研究所_光电信息技术研究室 |
| 推荐引用方式 GB/T 7714 | 于世孔. 自适应红外图像非均匀性校正方法研究[D]. 中国科学院沈阳自动化研究所. 中国科学院沈阳自动化研究所. 2015. |
入库方式： OAI收割
来源：沈阳自动化研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。