【作者】 刘伟；

【导师】 杨万海；

【作者基本信息】 西安电子科技大学 ， 电路与系统， 2003， 硕士

【摘要】 本文在分析了红外图像中背景、小目标和噪声特点的基础上，进行了两方面的研究。第一个方面是从红外图像中小目标的特性出发，对图像进行高通滤波，得到主要包含小目标和噪声的图像。然后，采用本文提出的基于小波变换中软阈值去噪原理和相关去噪原理的检测方法检测出小目标。同时，仿真结果表明基于软阈值去噪原理的算法优于基于相关去噪原理的算法。第二个方面主要是通过抑制背景来检测小目标。这一方面又分为两种情况。第一种情况是针对连续的多帧图像。分别采用时域中值滤波器算法和最小平方中值滤波器算法抑制背景，得到由小目标引起的变化的部分。第二种情况是针对单帧图像，采用数学形态学抑制背景，得到保留小目标的图像。然后，对包含小目标信息的图像进行处理，从而检测出小目标。同时，仿真结果表明，数学形态学算法最优，时域中值滤波器算法次之，最小平方中值滤波器算法最差。更多还原

【Abstract】 Based on the different characteristics of background, small targets and noise in infrared images, the research is conducted in two different ways in this paper. One is based on the characteristic of small targets. First, one image is filtered by a high pass filter and the image mainly including the small targets and noise are obtained. Then, using the algorithms with wavelet transformation presented in this paper, which are based on the soft-threshold denoising and the relevant denoising, the small target can be detected. At the same time, the simulation result shows that the performance of the algorithm that is based on the soft-threshold denoising is better than the algorithm that is based on the relevant denoising. The other is mainly depended on the background rejection to detect the small target. There are two different situations in this field. One aims at the image sequences. Using the LMedS (Least median of squares) algorithm and the temporal median filter algorithm independently, the background rejection can be realized, and the changed parts in the image sequences that are mainly caused by the small target are obtained. The other aims at one image. Using the mathematic morphology algorithm, the background rejection can be realized and the image mainly including the small target can be obtained. Then, the images that are obtained by the LMedS algorithm, the temporal median filter algorithm and the mathematic morphology algorithm are processed and the small target is detected. At the same time, the simulation result shows that the performance of the mathematic morphology algorithm is the best, the performance of the temporal median filter algorithm ranks second, and the performance of the LMedS algorithm is the worst.更多还原