A novel object-level change detection method based on multi-scale image segmentation

Abstract

Aiming to improve object fragmentation and poor detection results caused by discontinuous segmentation scales in object-level change detection, a new object-level change detection method based on multi-scale segmentation is presented in this paper. Firstly, a convexity model concept to describe target- background characteristics is proposed. This model is used to implement the convexity model-based multi-scale image segmentation, in order to overcome the shortcoming that traditional single-scale image segmentation can hardly synchronously extract the objects within different scales. And then, a change detection approach by analyzing structural characteristics of image objects is introduced, in order to detect the man-made object. Experiments show that the new method is robust and that it provides an advanced tool for quantitative change detection.