SAR图像显著性区域检测算法

发布时间：2018-04-26 12:01

本文选题：SAR图像 + 显著性区域检测　；参考：《西安电子科技大学》2015年硕士论文

【摘要】：随着合成孔径雷达(Synthetic Aperture Radar,SAR)成像技术的成熟和侦察范围的增加,现有的SAR图像数据量远超过了目前解译SAR图像所承受的能力,仅仅获取这些高分辨率的SAR图像并没有太多意义,重要的是解译和提取SAR图像中的重要信息。那么,如何从海量数据中分析并提取出这部分有用的信息就显得尤为重要,而视觉显著性区域检测的出现为解决这一问题提供了有效方案。由于乘性噪声的影响,使得SAR图像在显著性区域检测领域鲜有研究。因此,我们通过对光学图像中许多经典算法的研究,选择了LC(Linear-color Contrast)模型对SAR图像进行显著性区域检测。考虑到SAR图像与光学图像的不同成像机制,不可能将光学图像中的方法直接应用到SAR图像中,因此,我们需要在LC模型的基础上进行改进。下面介绍一下本文提出的两种基于LC模型的SAR图像显著性区域检测算法:第一种算法:基于局部相似度的SAR图像显著性区域检测算法。由于LC模型突出强调稀有颜色的特点,以及部分SAR图像颜色复杂度低、层次分明的特点,我们用LC模型对这一类SAR图像提取初级显著图。由于SAR图像乘性噪声的影响,初级显著图中存在大量噪声混合在显著区域。对此,我们在初级显著图的基础上,对目标区域所有像素计算其与周围像素的相似度之和。相似度越大,证明像素邻域内多为目标点,则判定该区域为显著区域;相似度越小,证明像素邻域内多为背景点,该像素存在于背景中,则判定该区域为背景区域。接下来,将初级显著图与相似图相乘,使得最终显著图中的显著区域得以增强,背景区域得以削弱,以此来减弱噪声的影响。第二种算法:基于超像素的SAR图像显著性区域检测算法。首先,用LC模型对SAR图像提取初级显著图。然后,用SLIC(simple linear iterative clustering)超像素分割算法对SAR图像进行超像素分割。计算每个超像素的平均灰度,并赋值给超像素中的每个像素点,使得比较尖锐的噪声与周围像素融合在一起。这样,经过聚类后的SAR图像,背景中噪声的影响明显减弱。此时,基于背景灰度均匀的特点,再运用LC模型对聚类后的SAR图像进行显著区域检测。最后,将该显著图与初级显著图相乘,最终也能获取较高质量的显著图。
[Abstract]:With the maturity of synthetic Aperture Radar (SAR) imaging technology and the increase of reconnaissance range, the existing SAR image data far exceeds the current ability to interpret SAR images, so it is not meaningful to simply obtain these high-resolution SAR images. It is important to interpret and extract important information from SAR images. So how to analyze and extract the useful information from massive data is particularly important and the emergence of visual salient region detection provides an effective solution to this problem. Due to the effect of multiplicative noise, there is little research on SAR images in the field of significant region detection. Therefore, through the study of many classical algorithms in optical images, we select LC(Linear-color Contrast-based model to detect the significant region of SAR images. Considering the different imaging mechanisms of SAR images and optical images, it is impossible to directly apply the methods in optical images to SAR images. Therefore, we need to improve the LC model. Two significant region detection algorithms for SAR images based on LC model are introduced in this paper. The first one is a significant region detection algorithm for SAR images based on local similarity. Because the LC model highlights the characteristics of rare colors, and some SAR images have low color complexity and distinct layers, we use LC model to extract the primary salient images of this kind of SAR images. Due to the effect of multiplicative noise in SAR images, a large number of noises are mixed in the significant region in the primary salient map. On the basis of the primary salience map, we calculate the sum of the similarity between the pixels in the target region and the surrounding pixels. The larger the similarity is, the more the pixel neighborhood is the target point, the more significant the region is, the smaller the similarity degree is, the more the pixel neighborhood is the background point, the more the pixel exists in the background, the more the region is the background region. Then, by multiplying the primary salience map with the similar map, the significant area in the final significant map is enhanced and the background area is weakened, so as to weaken the effect of noise. The second algorithm: SAR image salience region detection algorithm based on super pixel. Firstly, the primary salient map of SAR image is extracted by LC model. Then, the SLIC(simple linear iterative clustering algorithm is used to segment the SAR image. The average gray scale of each super pixel is calculated and assigned to each pixel in the superpixel to make the sharp noise merge with the surrounding pixels. In this way, after clustering SAR images, the influence of noise in the background is obviously weakened. In this case, based on the uniform gray level of background, the LC model is used to detect the significant regions of the clustered SAR images. Finally, by multiplying the salience map with the primary salience map, a higher quality salience map can be obtained.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN957.52

【相似文献】