【作者】 白俊奇；

【导师】 陈钱；

【作者基本信息】 南京理工大学 ， 电子科学与技术， 2010， 博士

【摘要】 近年来,红外成像技术取得了很大进步,在国防建设和国民经济各领域得到广泛的应用,如目标跟踪、成像制导和环境监测等。受探测器结构、制造工艺以及材料等因素限制,红外探测器存在非均匀性噪声和欠采样等问题,导致温度分辨率和空间分辨率不高,影响了红外系统的成像质量,限制了红外整机的探测距离,成为红外成像技术发展的瓶颈。高分辨率成像一直是红外成像领域研究的重点,不仅在理论上具有重要意义,实用中也有迫切需求。本文以提高红外成像温度分辨率和空间分辨率为目标,主要围绕红外成像的非均匀性校正和超分辨率重建展开,具体内容包括：1)红外焦平面阵列探测器非均匀性校正技术。从红外焦平面阵列、读出电路和环境温度三个方面讨论了非均匀性的产生机理,建立了探测器响应与环境温度的指数关系模型。针对探测器非均匀性随着时间漂移的问题,提出三种改进的场景非均匀性校正算法。结合红外焦平面阵列探测器输出信号的时间和空间特性,利用读出电路相邻通道差异的高斯分布,估计通道放大器引起的固定竖条噪声,提出将单帧迭代和黑体标定相结合的非均匀性校正算法。改进的神经网络非均匀性校正构造了基于读出电路的隐含层期望值,并应用LMS原理自适应调整步长以提高收敛速度。根据恒定统计理论,提出一种时域高通非均匀性校正中的“鬼影”抑制算法,将已估计偏置参数作为先验条件判断当前像素是否参与当前帧偏置参数的计算,解决了固定阈值场景适应性差的问题。2)微扫描超分辨率成像技术。研究了微扫描成像的图像处理技术,包括微扫描图像预滤波和图像插值。在图像预滤波部分,针对P-M方程解的病态性,提出改进的各向异性扩散滤波算法。算法结合像素梯度和局部方差构造了噪声低敏感性的扩散系数,该系数从理论上满足Charbonnier等提出的构造准则。在图像插值部分,提出改进的边缘保持图像插值和Lagrange多项式图像插值算法。边缘保持图像插值算法建立了局部标准差边缘模型,将图像分为平滑和边缘区域,平滑区域以像素邻域差值作为加权系数实现图像插值,边缘区域选用最大梯度方向的像素实现图像插值。Lagrange多项式图像插值算法将图像划分成多个的同性区域,在各个同性区域内自适应调整Lagrange的阶数,完成图像插值。3)多帧图像的超分辨率重建技术。围绕图像配准和图像重建问题展开研究。在图像配准部分,基于数学形态学和Mallat多分辨率分析理论,提出一种多尺寸模板配准算法,算法有效抑制了噪声影响；在图像重建部分,从Bayes随机场角度,基于Huber-Markov正则化模型,提出一种梯度阈值自适应调整的MAP超分辨率重建算法。实验测试和仿真表明,本文提出的技术与现有技术相比具有一定的优势。这些新技术已经应用到制冷和非制冷凝视型红外成像系统以及红外微扫描成像系统中,得到了满意的成像效果,展现了这些技术在提高红外图像质量方面的实用价值。更多还原

【Abstract】 Infrared imaging technique has achieved much progress and been applied to military and civilian fields in recent years, such as target tracking, imaging guidance, environmental monitoring and so on. Due to the structure of detector, manufacture technology, material etc, there are some problems in infrared detector, such as nonuniformity noise and undersampling, which result in low temperature resolution and spatial resolution, bad imaging quality and short detecting distance. High resolution imaging, which is theoretically important as well as practically urgent, is an active topic in infrared imaging region during recent years. For improving temperature resolution and spatial resolution of infrared imaging, nonunifomity correction and superresolution of infrared imaging are mainly discussed. The studies are carried out in the following aspects:1) Nonuniformity correction technique of infrared focal plane array (IRFPA) detector. The mechanism of nonuniformity was discussed from IRFPA, readout circuit and environmental temperature. The exponential relationship between detector response and environmental temperature was established. For solving the nonuniformity drift in time, three novel sence-based NUC algorithms were proposed. Aiming at the time and spatial features of signal output, a nonuniformity correction algorithm based on iterative theory and blackbody calibration was presented. The fixed stripe noise was estimated using Gaussian distribution of the difference of adjacent readout channels. New expectation estimation was constructed and the step value was adjusted adaptively based on the LMS theory in the improved neural network nonuniformity correction. A ghosting eliminated technique was proposed based on high-pass in time domain and low-pass in spatial domain. Accordding to constant-statistics theory, current pixel whether was used for computing offset parameter or not depending on the existed offset estimation.2) Microscan superresolution imaging technique. We discussed the image processing of microscan including pre-filtering and interpolate construction. In the pre-filtering part, aiming at ill-conditioned problem of the P-M equation, an improved anisotropic diffusion algorithm was presented. Based on pixel gradient and local variance characteristics, a diffusion coefficient was established which satisfies the condition of diffusion coefficient presented by Charbonnier. In the interpolation reconstruction part, an edge-preserved interpolation and an improved Lagrange interpolation algorithm were proposed, respectively. According to local root-mean-square edge model, the image was divided into smoothness and edge regions. The interpolation was carried out by using adjacent pixels in the smoothness region and the pixel of greatest gradient direction in the edge region, respectively. Improved Lagrange interpolation algorithm divided image into multi-isotropy regions, the exponent number of Lagrange equation was adjusted self-adaptively in every isotropy region.3) Multiframe superresolution reconstruction. Our discussion focuses on image registration and reconstruction. In the image registration part, a multi-scale templates registration algorithm was proposed based on morphological and Mallat multi-resolution analysis theories, which can effectively eliminate the noise. In the image reconstruction part, from Bayes random field angle, a self-adaptive threshold MAP superresolution reconstruction was proposed based on Huber-Markov model of regularization.Experimental tests and simulations show the proposed algorithms are superior compared with existed algorithms. These new technologies have been applied to the imported cool and uncool staring infrared imaging system and microscan imaging system, which achieve promising results. All of these show these key techniques have real value for improving the quality of infrared image.更多还原