Multi-resolution segmentation of soil moisture imagery by watershed pyramids with region merging

A new image segmentation method is presented for application to remotely sensed imagery. The method incorporates the advantages of previous watershed segmentation techniques while greatly improving the two main disadvantages of such techniques: computational efficiency and control over region scale. Computational efficiency is attained through the use of a multi-resolution image pyramid, and control over region scale is accomplished through a variational region merging algorithm. We apply the segmentation method to remotely sensed estimates of surface soil moisture. Experimental results of computational efficiency and comparison with other segmentation methods are presented.