Extraction of Change Information based on Multi-sequence Image Objects

Abstract

For identification of change information, most studies focus on pixel-based techniques for low-resolution images, while few studies have examined object-based techniques for high-resolution images. Moreover, most of the techniques are complex and have a high requirement for the segmentation scale. This paper proposes a change detection method based on multi-sequence image objects and introduces the use of arithmetic progression to generate the set of segmentation scales. Pre-event and post-event images are segmented with multi-scales, respectively, and sub-objects are obtained based on the division of the minimum segmentation scales of bi-temporal images. Change feature vectors are constructed for each associated object of sub-object and vectors' magnitude is computed. After the determination of change threshold values, the change feature vectors are used to confirm whether sub-objects have changed, providing final change information. This method was tested using the bi-temporal World View 2 images taken before and after a landslide. The results confirm the feasibility of the method presented in this paper, and show its high accuracy through a comparison with the changing vector analysis method and the post-classification comparison method based on object-oriented theory. The approach outlined herein would be helpful for extraction of change information in high-resolution images.