IEEE Geoscience and Remote Sensing Magazine - December 2019 - 15

the DSM, and seamline detection can then be cast as a
graph-cut problem [112]. With the help of external data,
the optimal seamline can be detected with a weakened
dependence on the image information itself. In brief,
external data can provide favorable ancillary information
for seamline detection.
Owing to the differences in radiation, illumination, projection, and so on, a reasonable way to obtain a seamless
mosaic is to find an optimal seamline for a smooth transition from one image to another neighboring image. For the
image internal information methods of seamline detection,
frame-to-frame methods are less efficient than multiframe
joint methods. Moreover, frame-to-frame methods usually
bypass the obvious ground objects, while multiframe joint
methods cannot. Thus, multiframe joint methods are often
accompanied by certain optimization strategies.

distance from the seamline. Typical weighting functions
are inverse distance weighting (IDW) and inverse cosine
distance weighting (ICDW) [3]. As shown in Figure 8, the
yellow rectangular region is the overlapped area of images a
and b. In a given buffer, d i denotes the distance from point
Pi of image b to the seamline. For a smooth transition between the buffer of the two images, this is often
f

P i = ~ ai P ia + ~ bi P ib,

where P ia and P ib are the pixels in Pi from image a and image b, respectively; ~ ai and ~ bi ( ~ ai + ~ bi = 1) are their corf
responding weights; and P i is the final mosaic result. The
weights are usually a function of the distance. As shown
in Figure 9, ICDW has a smoother effect in the edges (for
0 and 1) than IDW.
There are many types of weighting strategies. For multiresolution blending, the weighting strategy can be realized
with a pyramid [1], [2], [122], [123]. In addition to the image
domain, weighted blending is also suitable for the gradient

IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE

Weight

Weight

IMAGE BLENDING
Minor radiometric inconsistency can still exist around the
detected optimal seamline. To obtain a natural and smooth
transition from one image to another neighboring image,
image blending (also known as seamline elimination, feathering, or alpha blending) is usually required [87]. A large numRight Border
Left Border
ber of image-blending methods have been proposed.
j
0
Only a few methods have been suggested to get rid of
differences directly. Peleg [113] removed the seamline by
subtracting a smooth seam-eliminating function. Similarly,
Pi di
a hard correction method [114] can be used to compute the
Seamline
l
average gray difference between the pixels along the seamline and then adjust the gray difference to the pixels [89]. To
l
some degree, hard correction is similar to the linear transia
tion method [115], [116] and Poisson image editing [117],
b
[118]. Whitaker [16] proposed a level-set blending method
Buffer
by minimizing the difference metric on the level set rather
Overlapped Area
i
than the differences in pixel intensity values. The differences along the seamlines can also be reduced by a lowfrequency smoothing approach [8] or the use of a median
FIGURE 8. The buffer for image blending [3].
filter operation [45]. Generally speaking, direct elimination
along the seamline can reduce the radiometric differences effectively but
not completely.
1
1
To reduce the artifacts along a
seamline, a weighted combination of
0.8
0.8
the two neighboring images within
a transition zone is commonly used.
0.6
0.6
This combination can be achieved by
wa
wa
b
w
wb
the use of a bilinear weighting func0.4
0.4
tion [17], weighted averaging [12],
[71], [87], [119]-[121], or mean value
0.2
0.2
seamless cloning [18]. In other words,
each image is multiplied by a weight0
0
0
0.2
0.4
0.6
0.8
1
0
0.2
0.4
0.6
0.8
ing function in the transition zone
Distance Ratio (%)
Distance Ratio (%)
and then summed to form the final
(a)
(b)
mosaic. The core of this method is the
weighting function, which should
decrease monotonically with the FIGURE 9. The weighting functions [3]: (a) IDW and (b) ICDW.
DECEMBER 2019

(3)

1

15
IEEE Geoscience and Remote Sensing Magazine - December 2019

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - December 2019
