基于纹理信息的高分辨率无人机遥感图像分割

【作者】 武维；

【导师】 顾行发；

【作者基本信息】 电子科技大学 ， 地图制图学与地理信息工程， 2010， 硕士

【摘要】 虽然近年来高分辨率遥感影像数据呈现急剧性增长,但日益增长的数据需求与落后的影像分析技术之间的矛盾却越来越突出,造成的原因主要是落后的影像分析技术不能把原始的遥感影像数据转化为工程应用中所需的数据。图像分割是从图像处理到图像分析的关键步骤,在图像工程中占有重要的地位。所以有必要对图像分割相关的两个方面进行了一定的分析和改进。第一个方面是向对象影像分割算法分析和改进。在分析了面向对象的边缘检测和区域增长法两种面向对象影像分割方法的基础上,重点对区域增长法从两个方面进行了改进:第一,设计了新的增长规则;第二,增加了异质点去除环节。这样使算法减少过分割现象并在抗噪声方面也得到提升,最终使图像分割质量得到了有效的提升。第二个方面是对纹理信息提取技术的研究。对比分析了Tamura纹理和灰度共生矩阵两种纹理信息提取方法重点对灰度共生矩阵各参数对纹理特征的影响进行深入的研究。为了使通过共生矩阵能得到更合理的纹理特征,首先对10种纹理特征间的相关性进行分析,从而选出具有代表性的纹理。然后对开窗大小与各纹理特征的间的关系进行分析,从而为计算共生矩阵时的开窗大小选择提供依据。最后按照前面的研究结果把纹理特征和光谱特征结合起来对高分辨率无人机遥感影像进行了面向对象的分割,并进行了相应的定性和定量分析。定量分析方面与不考虑纹理信息的分割结果通过优度实验法进行了对比研究。结果表明考虑了纹理信息的面向对象分割能更真实有效的反映图像中地物目标的整体结构,为进一步有效的地物分类提供保障。更多还原

【Abstract】 The contradictions between the increasing need of Remote Sensing Image Data and Outdated image segmentation technology is becoming increasingly evident, altHough the amount of High-Resolution Remote Sensing data growing exponentially in recent years. The outmoded Image Analysis technologies can not convert Raw Remote Sensing Image Data to be useful data for application in Engineering is tHough to be the major reason. Image segmentation is a key step in image processing and image analysis. So, it is necessary to renew the knowledge of Relevant Technologies and make improve, which include texture analysis and image segmentation algorithm in this thesis.Study on Object-oriented Segmentation algorithm of Remote Sensing Image. Based on the analysis of the edge detection and Seeded Region Growing algorithm, a new algorithm was brought up, which is different from classical algorithm in two aspects:First, a new criterion of region growing is designed. Secondly, we propose a region merging rules based on the area and contrast, which can restrain the over-segmentation problem effectively and improve the anti-noise ability. Experiment results indicate that the method can improve the quality of image segmentation.Study on texture information extraction technologies. The Comparison of the two kind of Texture information extraction technology was carried out. Through the study on the correlation of the 11 texture feature and the effect of the window size on texture features quality, a scientific basis for choosing texture feature and the optimum window size was presented.According to the Research above, a texture image segmentation algorithm based on object-oriented which combines texture features and spectral characteristics was presented. In order to verify the accuracy of image segmentation, the new algorithm was applied to segmenting UAVRS images and a comparison between this approach and classical classification approaches has been carried out. The study conclusion shows that the new method is and more reliable comprehensive in reflecting the ground objects in the image.更多还原