【作者】 蒋苏蓉；

【导师】 冯刚；

【作者基本信息】 华中师范大学 ， 计算机应用技术， 2003， 硕士

【摘要】 图像分割是图像工程中一项基础而关键的技术。它是任何图像分析过程中首要的任务，因为接下来所要做的(如特征提取，目标识别等)都取决于图像分割的质量。近年来随着机器视觉、模式识别和基于内容的图像检索等技术的不断升温以及彩色图像的大量使用，图像分割特别是彩色图像的分割显示出越来越重要的地位。 本文在比较和分析现有彩色图像分割技术的基础上提出了一种改进的区域生长算法——色度空间上基于子块区域生长的彩色图像分割方法。颜色空间选取与人类视觉一致的HSV空间，在奇异点附近以非彩色信息作色调值的加权值。该方法不但考虑了像素的属性，还考虑了像素群的属性。图像首先被分成合适大小的子块，子块的色彩均值和方差值作为像素群的属性，用基于子块的区域生长来进行图像分割；根据提取出的对象大小以及它们的空间位置关系，去除掉过小的噪声对象同时将有意义的小对象合并到其所属的大对象中；最后处于边界的子块将逐像素地归类到对应的相邻对象中。 为了分析算法的性能，我们用多幅图像对其进行验证并对结果进行了分析。实验表明，该方法可以适合于多种类型的图像并且其计算量远小于传统的基于像素的区域生长方法，同时有助于改善由图像噪声所引起的过分割现象。更多还原

【Abstract】 Image segmentation is the basic and important technology in Image Engineering. It is a first task of any image analysis process because all the subsequent tasks (feature extraction and object recognition etc.) rely heavily on the quality of the segmentation. In recent years with the development of technologies, such as machine vision, pattern recognition and content-based image retrieve etc, and the wide use of color images, image segmentation, especially color image segmentation, has played a more and more important role.Many existing methods for image segmentation have been studied and compared in this thesis, and then an improved region growing algorithm is proposed-color image segmentation using chroma space based on block region growing. HSV color space that is consistent with human eyes is selected; achromatic information is used as the weigh on the hue value for strange pixels. Not only the pixel properties but also the pixel group properties are considered. Firstly an image is divided into most suitable blocks. For each block, the mean and variance values can be seen as the pixel group properties. The image segmentation is done with the approach of region growing based on block mean and variance. Considering the size of extracted objects and their relative position information, get rid of the "noise" objects and merge the meaningful fragmentary regions into their corresponding bigger ones. Finally all pixels in non-object blocks are classified into their corresponding adjacent objects.In order to analysis the performance of the whole algorithm, some pictures are tested and the segmentation results are analyzed. The experiments confirm thismethod is suitable for many kinds of images, its complexity of computation is less than traditional pixel-based methods, and also it can help do with the oversegmentation.更多还原