An adaptive scale estimating method of multiscale image segmentation based on vector edge and spectral statistics information

Scale computation for multiscale image segmentation has become one of the key scientific problems in urgent need to be solved in the field of geographic object-based image analysis (GEOBIA). Due to the complexity of High Spatial Resolution Remote-Sensing Imagery (HSRRSI) data itself and the scale distribution differences among geographic features, it is difficult to effectively design a global scale parameter model to guide parameters setting in large scale regions and automatically produce an acceptable segmentation result simultaneously. Utilizing the vector edge and spectral statistics information, an adaptively global scale computation method named Global Scale Computation with Vector Edge (GSCVE) has been developed for multiscale segmentation, which is firstly proposed and implemented on mean-shift segmentation algorithm as an example. The highlight of the GSCVE algorithm is that it can calculate global scale parameters for multiscale image segmentation adaptively. The validity of GSCVE algorithm was verified directly by taking GeoEye and QuickBird images as segmentation experiments sample data, respectively. In addition, comparing with the renowned eCognition (R) multiscale segmentation algorithm, the relative advantages of GSCVE algorithm with adaptive property and the concurrence segmentation results of large and small scale geographic features are illustrated by the visual evaluation experiments simultaneously.