Soft image segmentation model

With the spatial resolution of remote sensing data increases, the information of spatial features become more and more important for interpretation of remote sensing images. Therefore, object based image analysis methods receive more and more attentions. Image segmentation is the most important step for object based methods. Many segmentation algorithms have been developed, however, there are always uncertainties or errors in image segmentation. Unfortunately, such uncertainty information was largely neglected in previous studies. In this paper, a soft segmentation model aiming to describe the uncertainties in the segmentation procedure was developed based on the multi-resolution segmentation, which is a bottom-up approach and follows a pair-wise object merging process. At each merging step, the soft segmentation model calculates probabilities of several adjacent sub-objects merged into super-objects on the next level. By combining the probability at each merging step, the final probability of each pixel merged into the super-objects on the top level can be acquired. A case study of an IKON OS image was conducted for validating the effectiveness of the proposed model. The result shows that the soft image segmentation model is able to represent the reliability of the hard segmentation and the existence of mixed pixels on the borders of the adjacent objects. The uncertainty information of segmentation may help to further understand the segmented result.