融合颜色和纹理特征的彩色图像分割

• 作 者: 贾士军 王昆
• 机 构: 国核电力规划设计研究院勘测事业部,
• 出 处: 测绘科学
• 发表日期: 2014年12期
• 关键词: 水平集 纹理特征 彩色图像分割 建筑物提取 颜色特征 分割模型 Bayesian 高斯分布 高分辨率遥感 目标提取

根据高分辨率遥感影像中地物呈高斯分布的特性,文章将水平集与Bayesian统计理论相结合,提出一种统计理论框架下的广义水平集分割模型;并利用多元高斯分布融合图像的颜色和纹理信息,实现彩色图像的分割及建筑物目标提取。实验结果表明,该模型实现了与eCognition软件中多尺度分割相近的效果,且能够实现不同尺度目标的分割,易于向高维特征扩展,可扩展性强。

According to the characteristics of Gaussian distribution of high-resolution remote sensing images, this paper combines the level set with the Bayesian statistical theory to propose a generalized level set segmentation model under the framework of statistical theory. By using multivariate Gaussian distribution to combine the color and Texture information, the realization of color image segmentation and building object extraction. The experimental results show that this model achieves the similar effect as eCognition software in multi-scale segmentation, and can achieve the segmentation of different scale targets. It is easy to extend to high-dimensional features and has high scalability.