Journal of Sensors
from the gray values. e multiscale interest point detection
algorithms detect the scale of the image features and they are
strongly invariant to scale changes and other image artifacts
such as illumination variation, noise, blur, rotation, and ane
transformation.
Scale-invariant feature transform (SIFT) algorithm pro-
posedbyLowe[] is one of the most famous and popular
interest point detectors and has been proven to be robust in
many applications []. e SIFT interest points have a robust
repeatability property against image translation, rotation, and
scaling. Generally speaking, there are three major stages in
a typical SIFT algorithm. e rst stage is to detect extrema
in DoG scale space. Secondly, interest points are ltered and
located. e nal stage is to assign orientation and generate
descriptor for interest points. Even though a large number
of works have been devoted to improving SIFT, much eort
of most existing works in the literature has been spent in
focusing on the second and third stage, that is, improving the
description power of the descriptor for an interest point. Ke
and Sukthankar [] applied principal component analysis
(PCA) to gradient patch of an interest point and reduced
descriptor from D to D to form PCA-SIFT. e gradient
location and orientation histogram (GLOH), proposed by
Mikolajczyk and Schmid [], replaced the Cartesian location
gridusedbytheSIFTwithalog-polaroneandalsoapplied
PCA to reduce the dimension of the descriptor. Winder and
Brown [] made use of discriminative learning method to
optimize local descriptors under matching constraints from
a D construction. Abdel-Hakim and Farag [] extended
the SIFT algorithm to extract colored local invariant feature
descriptor, named as Color-SIFT. Verma et al. []proposed
new color-SIFT descriptors by extending the SIFT descriptor
to dierent color space. Li et al. []proposedanewlearning-
to-rank framework to reject unstable extrema such as low
contrast points and edge response points derived from the
rst stage of SIFT in order to improve interest point detection.
Liao et al. []presentedamodicationtotheSIFTinwhich
the normalized elliptical neighboring region was instead of
a rectangular region and a histogram was computed in a
polar space. By contrast, only limited work has been devoted
to solving problems in the rst stage, that is, identifying
locations and scales of potential interest points that can be
repeatably assigned under diering views of the same object.
In this paper, we apply Bilateral lter and Laplacian lter
on SIFT algorithm and present a novel method to detect
repeatable interest point with Laplacian-of-Bilateral (LoB)
lter. We rst smooth images by successively larger Bilateral
lter and a series of smoothed images are called scale space
or image pyramid. Compared with conventional Gaussian
lter, the Bilateral lter can remove noise from images while
preserving edges. Aerwards, we apply xed window Lapla-
cian lter on smoothed images in order to produce dierence
images called Laplacian-of-Bilateral (LoB) scale space. As a
consequence, scale space extrema detection can be imple-
mented eciently in a series of LoB images by convoluting an
image with LoB lter (with dierent scales). e next steps are
similar to standard SIFT algorithm for location, orientation
assignment, and descriptor generation.
e rest of the paper is organized as follows. In Section ,
we give a brief overview of the SIFT algorithm and explain
its drawbacks. en, our proposed interest point detection
algorithm is presented in Section . Experiment results are
discussed in Section .Finally,weconcludethispaperin
Section .
2. SIFT Algorithm Review
Before presenting in detail our approach, we briey review
the SIFT that forms the basis for our work. e SIFT
algorithm proposed in [] usually consists of three steps.
First, a Gaussian scale space is constructed and candidate
points are extracted by searching local extrema in a series
of DoG images. en all candidate points are localized to
pixel-accuracy and unstable points of low contrast or strong
edge response are eliminated. Finally, dominant orientation is
identied for each survived point, and its descriptor is gener-
ated based on the image gradients in its local neighborhood.
In order to provide a clear background for further
discussion, we give more insight into the details of the rst
step in SIFT. In the rst step, the scale space or image pyramid
is constructed by a variable-scale Gaussian, which is
,,=
1
2
2
exp
−
2
+
2
2
2
,
()
where (,)denotes the coordinate of a point and the scale
factor is .
Mikolajczyk found that the extrema of scale-normalized
Laplacian-of-Gaussian (LoG),
2
∇
2
, are the most stable
local points on an image compared to a range of other image
functions such as gradient, Hessian, and Harris. Moreover,
the dierence-of-Gaussian lter provides a close approxima-
tion to the scale-normalized LoG, which is
,,−,,=
(
−1
)
2
∇
2
,
()
where is a constant factor. So scale space extrema extracted
intheDoGfunctionconvolvedwiththeimage,(,,),are
regarded as candidate points, and
,,
=
,,
−
,,
∗
,
,
()
where (,)is an input image.
An eective way to construct (,,)and detect local
extrema is shown in Figure .einitialimageisincre-
mentally blurred with successively larger Gaussian lters to
produce images which are separated by a constant multiplica-
tive factor in scale space, shown stacked in the le column.
Aerwards, dierence-of-Gaussian images are produced by
subtracting each blurred image from the adjacent (more
blurred) image, shown on the right. For detecting the local
extrema (maxima or minima) of (,,), each sample
point is compared to its eight neighbors in the current image
andnineneighborsinthescaleaboveandbelow.esample
point is selected as a candidate point if its value is larger or
smaller than all of its neighbors.