【作者】 胡子付；

【导师】 曾志远；

【作者基本信息】 南京师范大学 ， 地图学与地理信息系统， 2006， 硕士

【摘要】 本文是在曾志远先生主持的国家自然科学基金项目“SPOT图像信息提取与土地利用／土地覆盖识别与监测研究”(批准号：40371053)资助下，对彩色图像主导颜色的提取及从属颜色的归并、归并后图像色斑边界的自动勾绘的理论和方法进行了研究。 目前遥感的应用可以分为两大类，一类是基于遥感影像多光谱特征的遥感影像的计算机分类；另一类是基于遥感彩色合成图像的人工解译，且该方法目前应用较广泛。为减轻基于人工解译遥感应用的工作量和减少解译过程中主观失误带来的解译结果的粗放性问题，本文就是要开发一套软件来代替基于图像颜色的人工解译。遥感影像彩色合成图像特别是基于曾志远先生LBV变换后图像的颜色一般与特定类型的地物相对应，遥感应用中一般只会提取最主要的若干种地物，对应到图像上就是提取最主要的若干种颜色，并把图像其他颜色归并到主要的颜色中去。 本文除对数字图像处理的一些基本理论问题进行研究外，重点对八叉树颜色量化算法进行了深入研究。发现利用八叉树算法归并的图像并不适合于地学应用，因为该算法用每一类所有像元颜色的平均值作为归并后类的颜色，当所要归并的类别数较少时，就会出现颜色的主色调与原图像发现较大的差异。本文对八义树算法进行了一些改进，用每一类所有像元颜色的优势色，即类中出现次数最多的颜色作为归并后类的颜色，取得了较好的颜色提取和归并效果；同时对现有GIS开发包的矢量化功能进行了研究。根据以上理论，开发出一套基于颜色对彩色图像进行归并和归并后图斑边界自动勾绘的软件系统。更多还原

【Abstract】 The Research is funded by the national natural science fund project titled by Information Extraction of SPOT Satellite Remote Sensing Image and Dynamic Monitoring of Land Use and Land Cover(sanction number:40371053).Method of color image classify by the color of the pixels and the method of Vector are studied.At present, the application of Remote Sensing Technology is comprised of two means, one is Computer Classification of Satellite images based on the multi-spectral bands features of the image, the other is the manual interpreter of multicolor images synthesized by Remote Sensing images. This method is applied broadly today. In order to decrease the workload of the method of manual interpreter and to reduce the imprecise problem resulted from the subjective muff in the process of manual interpreter. In this paper, the software which can replace the manual interpreter method of Remote Sensing images developed, in the multicolor images composed by Remote Sensing images especially the multicolor images transformed by LBV Transformation Method which is developed by Zhiyuan Zeng professor, one color is corresponding to the special ground features. But Minority ground features are extracted in the application of Remote Sensing .correspondingly the importance colors are extracted, and the other colors are merged into the importance colors.In this paper, expect for the basic theory of digital image process, the algorithm of Octree Quantization is deeply researched. So the problem that the color images processed by Octree Quantization are not suitable for geographic application is discovered, because the Octree Quantization the algorithm of Octree Quantization uses the average color of all the pixels in one class as the color of the class, so when the number of the class is small, the hue of the color images after processed will departure from the original images. In order to get suitable images, the improved algorithm of Octree Quantization is developed, which uses the majority color of one class as the color of the class. Additionally the method of vector is also studied, finally one software is developed based on the forward theory.更多还原