A Study of Optimal Seamline Determination for Aerial Image Mosaicking

Abstract

Image mosaicking is a fundamental step in composing of a set of aerial orthoimages Determining the optimal seamline is thus important in seamlessly merging the image patches Numerous seamline searching approaches have been proposed to solve this optimization problem in previous studies Most of these studies apply Dijkstra’s algorithm (DA) to identify a seam under the objective function of the shortest path However seamlines identified using this strategy may pass high-cost regions potentially resulting in seam artifacts To address this problem previous studies set a threshold in filtering out high-cost pixels prior to determining the seamline or constrain the seamline to pass certain vector paths Nonetheless proper threshold values are case sensitive and information of vector graphics requires non-trivial processes of data collection and fusion This study proposes a novel method with several extensions on both the objective function of the seamline optimization and the optimization solver Without thresholding and other geographic data the proposed method avoids seamlines that pass high-cost regions by searching for a seam under the objective function of a path with minimal average mismatch Considering the color transition and computational cost of image mosaicking a hierarchical structure is also applied to the seamline determination in our study Determining a seamline in a specific level of the image pyramid reduces computational cost and also generates a seam zone that softens color transition Thus applying these strategies can decrease the occurrence of seam artifacts caused by the discontinuity of high-cost regions Qualitative and quantitative analyses of various image sets indicate that the proposed method can determine near-optimal seamlines

Date of Award	2015 Aug 26
Original language	English
Supervisor	Chao-Hung Lin (Supervisor)