基于内容感知的遥感图像缩放技术研究与实现

发布时间：2018-01-30 03:44

本文关键词： 遥感图像敏感对象识别三角网格鱼眼变换图像缩放　出处：《北方工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着航空航天技术的进步,如何能够对遥感图像中的敏感区域进行无失真的有效缩放成了现在很重要的研究热点。机场和桥梁在交通运输方面具有重要的作用且一直是国防军事打击的重要目标。如何能够做到从全局的角度重点观察遥感图像中这两种对象的细节信息是军事打击以及研究的重要基础。本文提出一种提取对象敏感度特征并结合对象一般特性对多光谱图像中的敏感对象进行识别的算法;提出一种三角网格与鱼眼变换相结合的遥感图像缩放算法,以实现敏感区域的最优缩放。多光谱图像因其不同的采集方式,使其本身具有很多分类特征,对这些特征的提取可以有效的识别出图像中不同的对象区域。本文提出的一种基于敏感度的敏感区域识别算法,在对提取出的敏感对象潜在区域提取纹理特征、颜色矩特征、面积特征之后又结合各潜在区域在原始图像中的上下文信息提取了对象区域敏感度特征,最后,通过基于AdaBoost的SVM分类器对各潜在区域进行分类,确定敏感区域的具体位置,并在原图像中标识出来。将该算法应用于多光谱图像机场、桥梁敏感目标的识别上,实验表明该算法较以机场跑道直线特征为主要检测依据的机场识别算法和以桥梁与水体的光谱差异性为识别依据的桥梁识别算法的平均准确率有了明显提高,机场识别平均识别准确率达到92%,桥梁平均识别准确率达到了 90%。目前针对遥感图像缩放技术的研究非常少。本文提出的一种三角网格与鱼眼变换相结合的图像缩放算法针对已识别出的多光谱图像中的敏感区域,对区域内部进行三角网格划分,并对区域内部实现基于三角网格的缩放,保持区域内部敏感对象部分的信息内容不变。整幅图像除敏感区域之外的部分,做与敏感区域缩放过程产生的变化有关的基于鱼眼变换算法的压缩。实验结果表明,图像在缩放过程中产生的绝大部分失真畸变都发生在敏感区域以外的部分,敏感区域内部失真很小,缩放效果较现有的线裁剪缩放方法、三角网格缩放方法以及鱼眼变换内容感知缩放算法有明显改善。
[Abstract]:With the advance of aerospace technology. How to effectively scale sensitive areas in remote sensing images without distortion has become an important research hotspot. Airports and bridges play an important role in transportation and have always been important for national defense and military strike. How to observe the details of the two objects in remote sensing images from a global perspective is an important basis for military attack and research. In this paper, we propose a method to extract sensitivity features of objects and combine them with objects. The algorithm for the recognition of sensitive objects in multispectral images with general characteristics; A remote sensing image scaling algorithm based on triangular mesh and fish-eye transform is proposed to realize the optimal scaling of sensitive region. Multi-spectral images have many classification characteristics because of their different acquisition methods. The extraction of these features can effectively identify different object areas in the image. A sensitive region recognition algorithm based on sensitivity is proposed to extract texture features from the potential areas of the extracted sensitive objects. Color moment feature, area feature and the contextual information of each potential region in the original image are used to extract the sensitivity feature of the object region. The potential regions are classified by SVM classifier based on AdaBoost to determine the specific location of the sensitive regions and identify them in the original image. The algorithm is applied to the multi-spectral image field. Bridge sensitive target recognition. The experimental results show that the average accuracy of the algorithm is significantly improved compared with the airport recognition algorithm based on the linear features of the airport runway and the bridge recognition algorithm based on the spectral difference between the bridge and the water body. The average recognition accuracy of airport recognition is 92%. The average accuracy of bridge recognition has been achieved. At present, there is very little research on remote sensing image scaling technology. In this paper, a new image scaling algorithm based on triangle mesh and fish-eye transform is proposed for the sensitive areas in the identified multi-spectral images. Triangulation is carried out within the region, and the region is scaled based on the triangular mesh to keep the information content of the sensitive object in the region unchanged. The whole image is except the sensitive area. The experiment results show that most of the distortion in the process of image zooming occurs outside the sensitive region. The internal distortion of sensitive area is very small, and the scaling effect is obviously improved compared with the existing line clipping method, triangular mesh scaling method and fish-eye transform content sensing scaling algorithm.
【学位授予单位】：北方工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751

【参考文献】