Vegetation segmentation based on level set using local wavelet texture

Visible-spectrum indexes based methods for detection of vegetation coverage suffer from a dependence solely on color. To address these issues, this paper presents a graph cut and total variation based variational level set segmentation algorithm combining local wavelet texture and prior color. First, the prior color is generated through manual labeling with histogram estimation; local wavelet texture features are extracted using a multi-scale-and-orientation Gabor wavelet transformation followed by local median and entropy filtering. Second, in addition to the energy of color, that of texture is integrated into the variational level set model based on kernel density estimation. To obtain a global optimal solution, the convexity of the proposed energy functional is obtained using relaxation and threshold methods. Finally, a primal-dual algorithm with global relabeling is adopted to accelerate the evolution of level sets. A comparison of segmentation results between our algorithm and state-of-art ones is given. The experiments show that the presented algorithm can effectively reduce the dependence on color, and yields more accurate results for detection of vegetation coverage using remote sensing images.