An Efficient Multiscale SRMMHR (Statistical Region Merging and Minimum Heterogeneity Rule) Segmentation Method for High-Resolution Remote Sensing Imagery

Multiscale segmentation is an essential step for higher level image processing in remote sensing. This paper presents a new multiscale SRMMHR segmentation method integrating the advantages of Statistical Region Merging (SRM) for initial segmentation and the Minimum Heterogeneity Rule (MHR) for object merging. The high-resolution (HR) QuickBird imageries are used to demonstrate the SRMMHR segmentation method. The SRM segmentation method not only considers spectral, shape, and scale information, but also has the ability to cope with significant noise corruption and handle occlusions. The MHR used for merging objects takes advantage of its spectral, shape, scale information, and the local and global information. Compared with the Fractal Net Evolution Approach (FNEA) that eCognition adopted and SRM methods, the results show that the proposed method wipes off small redundant objects existed in traditional SRM methods, avoids the phenomena where the big homogeneity region has lots of small similar regions existed in the FNEA method, and gets more integrated and accurate objects. Therefore, the proposed SRMMHR segmentation method is an efficient multiscale segmentation method for HR imagery.