An improved seeded region growing-based seamline network generation method

Abstract

To generate an orthoimage product, mosaicking is a necessary process, and seam-based mosaicking of orthoimages is popular. However, many of these methods only focus on the generation of seamlines between two adjacent orthoimages, so the final generated mosaicking image depends on the order of compositing. To address this shortcoming, this paper presents an initial seamline network generation method based on improved seeded region growing. The basis of this method is the use of raster data rather than vector calculation, which is used with the area Voronoi diagrams with overlap (AVDO)-based method. First, the effective area of each image and overlap regions between adjacent images are determined. Then, the improved seeded region growing algorithm obtains the seamlines of each overlap region. The main improvement is that the boundary lines of overlap regions, rather than individual points, are chosen as seeds of the seeded region growing algorithm. These seeds grow simultaneously until growing regions overlap. The generated separatrix of growing regions is regarded as the seamline in the overlap region. At the same time, the cut result of the image's effective area is obtained. After that, these generated cut images are intersected to generate the effective mosaic polygon (EMP) of the image. Finally, all generated EMPs are vectorized to form the initial seamline network. In this way, the proposed method can process any kind of overlap region, and the final generated seamline network has no relation to the order of the image compositing. The experimental results demonstrate that the presented method is feasible and can achieve higher accuracy than the previous AVDO-based method.