基于视觉显著性的红外小目标检测算法研究

发布时间：2018-06-03 19:39

本文选题：视觉显著性 + 复杂背景　；参考：《华中科技大学》2016年硕士论文

【摘要】：近年来,显著性研究被成功地应用在目标检测、识别、图像压缩等多个领域,这些领域中就包括了红外小目标检测。作为红外自寻的制导、搜索跟踪和预警等领域的一项关键技术,红外小目标检测,特别是复杂云干扰下的空中目标、海杂波与云层干扰下海天背景的红外小目标、以及复杂地面背景下小目标检测依然是目前的研究热点和难点。本文在全面总结前人工作的基础上,对复杂背景下红外小目标图像的特征提取、视觉显著性模型的构建等展开了深入研究。论文的主要工作如下:首先,介绍了红外小目标图像的数学模型,分析了图像目标、背景和噪声的特性,通过预处理增加了背景与目标的对比度、抑制了图像的噪声。其次,在红外小目标图像数学模型的指导下,提出了一种新的基于频谱残差法的显著目标检测算法,该方法将经典的频域显著性方法与两个一阶梯度方向特征结合,提高了算法的抗噪性和准确性。然后,针对频谱残差方法存在的缺陷,提出了一种基于超复数幅度谱的红外小目标检测算法,研究了幅度谱与重复的非显著性区域的对应关系,通过结合图像二阶方向导数特征与灰度特性,将红外图像重构为超复数矩阵形式,进而使用高斯核对超复数傅里叶变换后的幅度谱进行滤波,抑制非显著性区域,使得目标这一显著性区域获得增强效果。经过实验验证,该算法具有更强的目标增强与背景抑制能力。最后,对于显著性图的分割问题,提出了两种思路,一种是传统的基于阈值分析的方法,另一种是基于模糊度量的阈值分割方法,实验结果证明了前一种算法更高效,准确性也有保证。
[Abstract]:In recent years, the significance research has been successfully applied in many fields, such as target detection, recognition, image compression and so on. These fields include infrared small target detection. As a key technology in infrared homing guidance, search tracking and early warning, infrared small target detection, especially in the sky under complex cloud interference, sea clutter and cloud interference, is a key technology in the field of ocean and sky background. And small target detection in complex ground background is still a hot and difficult point. On the basis of summing up the previous work, this paper makes an in-depth study on the feature extraction of infrared small target images and the construction of visual salience model in complex background. The main work of this paper is as follows: firstly, the mathematical model of infrared small target image is introduced, and the characteristics of image object, background and noise are analyzed. The contrast between background and target is increased by preprocessing, and the noise of image is suppressed. Secondly, under the guidance of the mathematical model of infrared small target image, a new significant target detection algorithm based on spectral residual method is proposed, which combines the classical frequency-domain saliency method with two first-order gradient directional features. The noise resistance and accuracy of the algorithm are improved. Then, aiming at the defects of spectrum residual method, an infrared small target detection algorithm based on hypercomplex amplitude spectrum is proposed, and the corresponding relationship between amplitude spectrum and repetitive non-significant region is studied. The infrared image is reconstructed into a hypercomplex matrix form by combining the second-order directional derivative and gray characteristics of the image, and then the amplitude spectrum of the super-complex Fourier transform is filtered by using Gao Si check to suppress the non-significant region. Increases the target's significant area. Experimental results show that the algorithm has stronger ability of target enhancement and background suppression. Finally, for the segmentation of salient graph, two methods are proposed, one is the traditional method based on threshold analysis, the other is the threshold segmentation method based on fuzzy metric. The experimental results show that the former algorithm is more efficient. Accuracy is also guaranteed.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【参考文献】