Optimal seamline detection for SAR image mosaicking guided by superpixel segmentation and region merging

Abstract

Large-scale studies and applications of SAR images require the mosaicking of multiple scenes. However, geometric misregistration and radiometric inconsistencies among adjacent images often lead to poor continuity and unnatural transitions in the mosaicked images, which severely restrict the effectiveness of SAR images in large-area information analysis and retrieval. Optimal seamline detection seeks to determine the most suitable stitching path within the overlapping regions of adjacent images, ensuring that mosaicked SAR images exhibit maximal consistency in intensity, texture, and geometric features while minimizing stitching artifacts and visual discontinuities. Existing seamline detection methods, however, are often limited by their obstacle-avoidance capability and computational efficiency. To overcome these limitations, this study proposes an optimal seamline detection approach for SAR images guided by superpixel segmentation and region merging. First, the Patch-Based SLIC (PB-SLIC) algorithm is enhanced to achieve consistent superpixel segmentation across multiple overlapping images. Second, a region adjacency graph is constructed by integrating Bhattacharyya distance, texture distribution, and boundary length information, which guides the merging of superpixels and produces candidate seamlines that preserve object integrity and accurately follow object boundaries. Then, an initial seamline network is generated using area Voronoi diagrams with overlap (AVDO), and a cost function based on normalized cross-correlation is established. The seamline network is further refined using a shortest-path algorithm to extract the optimal seamline network from the candidates. Finally, using real SAR datasets, we analyze and verify the effectiveness of superpixel segmentation and region merging in seamline detection. Comparative experiments with two classical methods further demonstrate that the proposed approach achieves superior obstacle-avoidance capability and shorter search time, while ensuring higher mosaicking quality and significantly improving computational efficiency.