Automated hierarchical segmentation of high-resolution remote sensing imagery with introduced relaxation factors

Abstract

This paper proposes a new automated hierarchical segmentation method with introduced relaxation factors for processing high-resolution remote sensing imagery, which aims to provide a theoretical framework in setting the scale parameters and reducing the influence of human factors. The first relaxation factor is used to adjust the heterogeneity between the image-objects to be merged, thus improving the speed of the entire segmentation by controlling the number of image-objects in each recursive merging. With the mean of the heterogeneity between image-objects taken as the cardinality, the second relaxation factor is introduced to control the scaling parameter of the levels exported in the process of segmentation, automatically producing multi-scale hierarchical segmentation results. The experimental results show that this method produces segmentation with higher quality, which meets the accuracy requirements of further image analysis and geographic object extraction. Other theoretical and practical contributions of this method include reducing the influence of human factors and improving the level of automation in segmentation. Further investigation is still required with respect to processing the boundaries of geographic objects with complex image, and increasing the compactness and smoothness of image-objects.