Sketch-based Image Retrieval Using Contour
Segments
Yuting Zhang
#1
, Xueming Qian
*2
, Xianglong Tan
#3
#
SMLESLAB of Xi’an Jiaotong University, Xi’an CN710049, China
1
zhangyuting@stu.xjtu.edu.cn
2
qianxm@mail.xjtu.edu.cn
3
xjtuicemaple@sina.com
Abstract—The paper presents a sketch-based image retrieval
algorithm. One of the main challenges in sketch-based image
retrieval (SBIR) is to measure the similarity between a sketch
and an image in contour with high precision. To tackle this
problem, we divided the contour of image into two types: the first
is global contour, suggesting that we can use it to reduce the
similarity between the images with complex background. The
second, called salient contour, is helpful to retrieve images with
objects similar to the query. Besides, we propose a new
descriptor, namely angular radial orientation partitioning
(AROP) feature, which makes full use of the gradient orientation
information to decrease the gap between sketch and image. Using
the two contours as candidate contours for feature extraction
could increase the retrieval rate dramatically. Finally an
application of retrieval system based on this algorithm is
established. The experiment on 0.42 million image dataset shows
excellent retrieval performance of the proposed method and
comparisons with other algorithms are also given.
I. INTRODUCTION
Developments in Internet and mobile devices have
increased the demand for powerful and efficient image
retrieval tools. Content-based image retrieval (CBIR) mainly
uses the text or an image as a query. Text features are less
accurate and might take mismatch between the user’s
expression and the user’s expectation. Although the image-
based search technology develops rapidly and works well,
there are some trouble in obtaining relevant images when the
user does not have the query images and text. To avoid this
problem, the user could draw a sketch and then use the sketch
as the input for an image retrieval system, this becomes more
and more convenient for users. Sketch-based image retrieval
(SBIR) technology becomes an active research area.
SBIR methods use a hand-drawn sketch composed of rough
and simple black and white to retrieve the corresponding
images. Although SBIR had been studied since 1990s, it still
remains challenge to measure the similarity between a sketch
and an image with high precision. Image retrieval must deal
with the ambiguousness in the query sketch caused by a lack
of semantic, besides a large majority of potential users fail to
precisely express fine details in their drawings [14, 15, 16]. To
improve the precision, many descriptors are proposed. Thus
many studies have been focussed on how to choose a good
descriptor. Some works focus on global descriptors, but the
other works focus on local descriptors. Some researchers
design a robust global descriptor to represent the sketch and
image individually. Global features can be better used in
image analysis, matching, and classification, such as HOG
(histogram of gradients) [1], EHD (edge histogram descriptor)
[2], and ARP (angular radial partitioning) [3]. However,
global features are unsatisfactory as they are unreliable under
affine variations. To overcome such drawbacks, Eitz et al. [4],
[5] use local descriptors to achieve state-of-art performance.
And QVE (query by visual example) [6] is a typical method
using blocks and local features. Cao et al. also propose a local
feature method, edgel index method [7], for sketch-based
image search by converting a shape image to a document-like
representation.
In order to develop an image retrieval system which is able
to find out more images with objects similar to the query, we
develop a global feature based on the global and salient
contours. The global contour is a global feature, and is defined
to find the relevant image with simple background. The salient
contour is local feature, and is defined to tackle the problem
that one object is similar to the query. Besides, the AROP
feature is refined the ARP feature, and makes full use of
contour orientation to constrain the shape information.
The main contributions of this paper are summarized as
follows. 1) We propose the global contour, which introduces
the salient region to make the contour more discriminative. 2)
We propose the salient contour to make the retrieve images
with objects similar to the query. 3) The orientation
partitioning scheme is introduced based on the original ARP
feature. Thus AROP feature contains more information, which
makes the retrieval result more accurate and reliable.
The remainder of this paper is organized as follows. Work
related to sketch-based retrieval is reviewed in Section II. We
describe the proposed approach in Section III, our
experiments in Section IV, and the discussion in Section V.
Finally, we present our conclusions in Section VI.
II. R
ELATED WORK
There have been a lot of studies in sketch-based image
retrieval system recently and sketch based image retrieval
techniques have been well discussed in [18]. In the following,
we briefly describe some approaches which are widely used in
SBIR system.
2015 IEEE 17th International Workshop on Multimedia Signal
Processing (MMSP), Oct 19-21, 2015, Xiamen, China.
978-1-4673-7478-1/15/$31.00 ©2015 IEE