Seamline Determination for High Resolution
Orthoimage Mosaicking Using Watershed
Segmentation
Mi Wang, Shenggu Yuan, Jun Pan, Liuyang Fang, Qinghua Zhou, and Guopeng Yang
Abstract
Image mosaicking is a process during which multiple orthoim-
ages are combined into a single seamless composite orthoim-
age. One of the most difficult steps in the automatic mosaicking
of orthoimages is the seamline determination. This paper pres-
ents a novel algorithm that selects seamlines based on mark-
er-based watershed segmentation. A representative seamline is
extracted at the object level and the pixel level as follows. First,
a watershed segmentation is performed to obtain the objects.
To avoid over-segmentation, a regional adaptive marker-based
watershed segmentation is proposed. Second, the object differ-
ence estimated by the correlation coefficient of each object is
calculated, and the region adjacency matrix is built. Third, a
technique for minimizing the maximum object cost is adopted
to determine the objects through which the seamlines pass.
Finally, pixel-level optimization is performed using Dijkstra’s
algorithm with a binary min-heap to determine the final seam-
lines. The experimental results on digital aerial orthoimages in
different areas demonstrate the feasibility and effectiveness of
the proposed method compared with other algorithms.
Introduction
Orthoimages integrate the rich information content of images
and the geometric properties of maps (ground projection) and
can be easily combined with additional information from
geographic information systems (
GIS
) to create an orthoimage
map. Thus, they have become a popular visualization product
and planning instrument (Kerschner, 2001).
Image mosaicking is defined as a necessary process in which
multiple orthoimages are combined into a single seamless com-
posite orthoimage for many applications, such as environmen-
tal monitoring and disaster management (Pan and Wang, 2011).
When mosaicking orthoimages, seamline determination is one
of the key steps and has an important impact on the quality
of the final mosaic. It is the process of finding a path with the
least difference in the overlapping area of the images to be
merged. The purpose of seamline determination is to minimize
as much as possible the discontinuity in the final mosaic of
the path. In an orthoimage, objects not included in the Digital
Terrain Model (
DTM
), or wrongly modeled, would appear at dif-
ferent locations in various orthoimages because of the variation
in viewing angles (Kerschner, 2001). Particularly for high-res-
olution aerial orthoimages of urban areas, different facets of an
object, e.g., a building, may appear in different images.
In previous studies, scientists attempted to develop meth-
ods to determine an optimal seamline to improve the quality
of orthoimage mosaicking. Milgram (1975) defined the “best”
seamlines by minimizing the amount of artificial edge in a
certain width in terms of pixel value difference between two
overlapping images to determine the lowest cost path. Fer-
nandez
et al.
(1998) defined seamlines by bottleneck shortest
paths. Fernandez and Marti (1999) developed a Greedy Ran-
domized Adaptive Search Procedure (
GRASP
) to optimize the
bottleneck shortest paths. Afek and Brand (1998) determined
seamlines based on a matching algorithm and local transfor-
mation. Kerschner (2001) proposed a “two snake” method to
find seamlines in areas with high color and texture similarity.
The two snakes approach the desired line in the image from
opposite sides. During their evolution, the two snakes attract
one another and the optimal seamlines are determined when
the two snakes merge into one. Soille (2006) proposed a mor-
phological image compositing algorithm. The main idea of the
algorithm was to allow for the automatic positioning of the
mosaic seamlines along salient image structures to diminish
their visibility. Chon
et al.
(2010) used Dijkstra’s algorithm
(Dijkstra, 1959) based on the Normalized Cross Correlation
(
NCC
) cost function, and limited the maximum difference in
seamline selection. This method enabled the search to possi-
bly find a longer seamline with less highly mismatched pairs.
Ma and Sun (2011) proposed an improved A* algorithm com-
bining lidar point clouds for orthoimage seamlines optimiza-
tion. This method was able to intelligently bypass the obstacle
areas smartly, and the resulting path could get close enough to
the initial seamlines. Yu
et al.
(2012) presented an automatic
seamlines location algorithm which used image appearance
(i.e., color, edge, and texture), image saliency and location
constraints for mosaicking optical remote sensing images.
Wang
et al.
(2012), Wan
et al.
(2012) and Wan
et al.
(2013)
suggested an approach using vector roads to aid in generating
seamlines for aerial image mosaicking. Pan
et al.
(2014b) put
forward a method of seamline determination by mean shift
segmentation for orthoimage mosaicking in an urban area.
In this paper, we present a new method of seamline de-
termination based on marker-based watershed segmentation
for high-resolution orthoimage mosaicking. It determines the
optimal seamlines at both the object and pixel level. First, the
Mi Wang and Jun Pan are with the State Key Laboratory of
Information Engineering in Surveying, Mapping and Remote
Sensing, Wuhan University, 129 Luoyu Road, Wuhan, Hubei
430079, China; and the Collaborative Innovation Center of
Geospatial Technology, 129 Luoyu Road, Wuhan, Hubei
430079, China (
).
Shenggu Yuan, Liuyang Fang, and Guppeng Yang are with the
State Key Laboratory of Information Engineering in Survey-
ing, Mapping and Remote Sensing, Wuhan University, 129
Luoyu Road, Wuhan, Hubei 430079, China.
Qinghua Zhou is with the China Railway Engineering Cosulting
Group Co., Ltd, 15 Guangan Road, Beijing100055, China.
Photogrammetric Engineering & Remote Sensing
Vol. 82, No. 2, February 2016, pp. 121–133.
0099-1112/16/121–133
© 2015 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.82.2.121
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
January 2016
121
