高分辨率遥感影像的全变分分割模型

Segmentation of high spatial resolution remote sensing image with Total Variation model

由于全变分(Total Variation,TV)模型具有较好的去噪、增强和扩散等功能,在过去的几十年中,TV模型在图像去噪、增强和超分辨率重建等方面得到了深入研究与广泛应用。鉴于TV模型的理论与分割理论具有一致性,因此本文主要研究TV模型用于高分辨率遥感影像的分割,并针对地物多尺度特征,提出了自适应的TV(ATV)模型;且与目前流行的面向对象的影像分析软件eCognition中的FNEA分割方法进行了比较。实验采用2幅高分辨率遥感影像,同时采用了面向对象的分割和分类评价,得出各方法各具优缺点的结论。

The total variation （TV） model is an effective tool for images processing such as image restoration, enhancement, reconstruction and diffusion techniques. Due to the consistence between the total variation model and the segmentation problem, in this paper, a TV-based segmentation approach was investigated for high spatial resolution remote-sensing imagery. Specifically, an adap- tive TV （ATV） model was proposed considering the multiscale characteristics of objects in high-resolution imagery. In experiments, the proposed TV-based approach was compared with the widely used Fractal Net Evolution Approach （FNEA） that is embedded in the commercial software eCognition.