QQ客服 官方微博 反馈留言

题名	激光对光电系统图像干扰的效果评估
作者	钱方
学位类别	博士
答辩日期	2015-05
授予单位	中国科学院大学
导师	郭劲
关键词	图像质量评价 激光干扰 小波变换 结构相似度 光斑特性
其他题名	Assessment of Laser-dazzling effects on Electro-optical System
学位专业	光学工程
中文摘要	光电成像系统主要利用各类光电探测器，通过对观测区域进行高分辨率成像获取目标及感兴趣场景的图像，提供给观察人员或图像处理系统对目标进行探测、跟踪与识别等应用。激光因其单色性、相干性、方向性、高亮度等优点被广泛应用于对光电探测器的辐照干扰与损伤。随着激光对光电探测器持续辐照功率的增强，会造成器件像元饱和甚至损伤，带来输出图像对比度降低等质量下降影响，造成使用者或处理系统无法提取、跟踪或识别目标。因此，迫切需要对激光干扰光电系统的图像降质原因展开分析，将人类视觉特征、干扰光斑特征、图像统计特征和局部特征损失等因素以定量方式进行描述，设计合理的评估方法，开展实验分析与改进，为激光干扰设备的研制与光电成像系统的激光辐照防护提供有效的参考依据。 研究过程中，针对提供给操作人员进行主观判别分析的成像侦察系统，提出了人类视觉系统特性与小波变换、图像特征损失和光斑性质变化相结合的三类评价算法。针对进行目标检测、跟踪与识别的图像处理系统，提出了基于特征点信息变化的评价算法。然而以上方法仅限于评价单帧静态图像，无法衡量一段时间内多帧图像激光干扰的综合累积效果，针对这类问题，提出了基于连续多帧动态特征变化的评价算法。通过大量实验分析，以上方法均能定量准确反映激光干扰效果，评价结果符合主观感知。本论文具体包括以下几个方面的内容： (1)理论分析部分 对于成像侦察系统，其终端用户是操作人员，所以人眼对图像质量的主观感知就反映了图像质量的变化程度。针对该类系统，本文提出了基于小波变换的多尺度加权结构相似度图像质量评价方法(WWMS-SSIM)，该方法模拟了人类视觉系统对图像处理的滤波过程以及对不同频带信息的敏感程度，使评价结果与主观感知比较一致。针对激光光斑对图像中信息的掩盖效应、图像特征的损失程度，以及实际试验条件下，原始图像很难获取的问题，提出了基于光斑特性加权的结构相似度的全参考图像质量评价方法(WFSIM)和特征变化度无参考图像质量评价算法(FVM)，这两种算法在分析人类视觉系统特性的基础上，考虑了光斑的大小、强度和位置等特性对图像质量的影响，实现了光斑信息变化与图像信息损失的结合，突出了激光干扰图像的特点，评价结果能更真实的反映干扰效果。 对于图像处理系统，是否能够准确提取图像特征反映了图像质量的变化程度。针对该类系统，提出了基于特征点分布特性的全参考图像质量评价方法(FPSIM)和特征点复杂度无参考图像质量评价方法(NRFPCM)，这两种方法考虑了干扰发生前后特征点数量、位置等信息的变化以及特征点周围亮度、对比度等图像信息的变化，使评价方法与目标识别准确性相结合，能够准确反映图像质量变化对目标识别的影响。由于激光干扰通常是一种连续变化的累积过程，仅对单帧干扰图像进行评估无法反映整体干扰效果，所以需要对多帧图像的质量变化加以评估。针对这类问题，提出了基于连续多帧图像动态特征变化的无参考图像质量评价方法(FPD)，算法不但可以评价单帧图像的质量，还可以分析多帧图像的综合干扰累积效果。 (2)实验仿真分析部分 算法有效性验证方面，在实验中通过对LIVE数据库中不同失真类型的图像进行质量评价，使用VQEG制定的评估标准，计算了线性相关系数、均方根误差数和误点率，验证了提出算法的准确性和一致性；算法适应性验证方面，使用NMSE、PSNR、SSIM和本文算法对不同背景强度、干扰功率和光斑位置的三类激光干扰图像进行质量评价，并将它们的评价结果进行对比，验证了提出算法的主观一致性和客观合理性；算法可扩展性验证方面，对部分算法的时间开销进行了分析，验证了算法在处理增量数据时的应用前景。
英文摘要	The photoelectric imaging system fulfils various functions such as high-resolution imaging, detection, tracking and recognition by means of photodetector. Laser is widely used in radiation jamming and damage because of its monochromaticity, coherence, directivity and high intensity. The imaging device forms are saturated and damaged with increasing of irradiance. This leads to quality reduction of the image and errors of the recognition. Therefore, it is necessary to understand the cause of distortion and analyze the human visual system characteristics, laser spot features, image statistical characteristics and the loss of location information quantitatively. The evaluation algorithm designed gives the guidance to the laser-dazzling equipment development and photoelectric imaging system defense. For the imaging reconnaissance system based on artificial cognition, the algorithms, combining with human visual system characteristics, by means of wavelet decomposition, losses of image features and varieties of laser spot are proposed. For the image processing system to fulfill target detection, tracking and recognition, the algorithms based on feature-point change are proposed. In order to evaluate the dazzling effects in the period of time, the algorithm based on the dynamic features of multi-frame images is proposed. In terms of experiment results，these algorithms give accurate and quantitative evaluation results for laser-dazzling images. The results are suitable for the subjective visual feeling. The mainly work studied in the thesis is arranged as follows: (1) Theoretical results and analysis For the imaging reconnaissance system, whose end user is observers, the subjective feeling reflects the variation of image quality. A wavelet-weighted multi-scale structural similarity (WWMS-SSIM) algorithm is proposed. This algorithm, based on the human visual system, simulates the image processing and sensitivity to different frequency bands. The results are consistent with the subjective feeling. Considering the masking effect and information loss of the laser-dazzling image, a full-reference weighted feature-similarity (WFSIM) algorithm and a no-reference feature variation metric (FVM) are proposed. These algorithms analyze the human visual system characteristics and consider the size, intensity, position of the laser spot. The algorithms combine laser spot feature change with image information loss. The evaluation results can reflect more reasonable dazzling effects. For the image processing system, the image feature extraction accuracy reflects the variation of image quality. A full-reference feature-point similarity (FPSIM) algorithm and a no-reference feature-point complexity metric (NRFPCM) are proposed. These algorithms analyze the position, amount of the feature points, local luminance and contrast distortion. The algorithms combine target recognition with evaluation methods. The evaluation results can reflect the image distortion impact on target recognition. Laser jamming is a continuous, variable and cumulative process, so the assessment of single frame cannot reflect dazzling effects. It is necessary to evaluate multi-frame image quality. A no-reference image quality assessment algorithm based on the dynamic features of multi-frame images is proposed, which is called feature-point dynamic (FPD). This algorithm can evaluate the laser-dazzling image quality of a single image and it gives a more reasonable evaluation result for multi-frame laser-dazzling effects in the period of time. (2) Experiment and analysis In the way of availability, we evaluate the image quality of the LIVE database. In order to get quantitative measures on the performance of these algorithms, we followed the performance evaluation procedures employed in the video quality experts group (VQEG). The correlation coefficients, root mean square error, mean absolute error and outlier ratio are calculated to verify the accuracy and consistency. In the way of adaptability, NMSE, PSNR, SSIM and algorithms proposed are used to evaluate different laser-dazzling images，containing different interference power, different background intensity and different spot position. The quality scores are compared and the subjective consistency, objective reasonableness are verified. In the way of expansivity, we analyze the time overhead of some algorithms. The applied prospect of dealing with incremental datas is verified.
公开日期	2015-12-24
内容类型	学位论文
源URL	[http://ir.ciomp.ac.cn/handle/181722/48887]
专题	长春光学精密机械与物理研究所_中科院长春光机所知识产出
推荐引用方式 GB/T 7714	钱方. 激光对光电系统图像干扰的效果评估[D]. 中国科学院大学. 2015.

个性服务

查看访问统计

相关权益政策

暂无数据

收藏/分享

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。