The Demo Site

This report is written by JIFENG NING,LEI ZHANG and DAVID ZHANG, CHENGKE WU

A novel object tracking algorithm is presented in this paper by using the joint colortexture
histogram to represent a target and then applying it to the mean shift framework.
Apart from the conventional color histogram features, the texture features of
the object are also extracted by using the local binary pattern (LBP) technique to
represent the object. The major uniform LBP patterns are exploited to form a mask
for joint color-texture feature selection. Compared with the traditional color histogram
based algorithms that use the whole target region for tracking, the proposed algorithm
extracts effectively the edge and corner features in the target region, which characterize
better and represent more robustly the target. The experimental results validate that
the proposed method improves greatly the tracking accuracy and efficiency with fewer
mean shift iterations than standard mean shift tracking. It can robustly track the target
under complex scenes, such as similar target and background appearance, on which the
traditional color based schemes may fail to track.


Introduction :

Real-time object tracking is a critical task in computer vision applications. Many
tracking algorithms have been proposed to overcome the difficulties arising from
noise, occlusion, clutter and changes in the foreground object or in the background
environment. Among the various tracking algorithms,mean shift tracking algorithms
have recently become popular due to their simplicity and efficiency.

The mean shift algorithm was originally proposed by Fukunaga and Hostetler
for data clustering. It was later introduced into the image processing community by
Cheng. Bradski odified it and developed the Continuously Adaptive Mean Shift
(CAMSHIFT) algorithm to track a moving face. Comaniciu and Meer successfully
applied mean shift algorithm to image segmentation and object tracking. Mean
Shift is an iterative kernel-based deterministic procedure which converges to a local
maximum of the measurement function with certain assumptions on the kernel
behaviors. Furthermore, mean shift is a low complexity algorithm, which provides
a general and reliable solution to object tracking and is independent of the target
representation.

The texture patterns, which reflect the spatial structure of the object,
are effective features to represent and recognize targets. Since the texture features
introduce new information that the color histogram does not convey, using the joint
color-texture histogram for target representation is more reliable than using only
color histogram in tracking complex scenes. The idea of combining color and edge for
target representation has been exploited by researchers.7,10 However, how to utilize
effectively both the color intensity and texture features is still a difficult problem.

This is because though many texture analysis methods, such as gray concurrence
matrices9 and Gabor filtering, have been proposed, they have high computational
complexity and cannot be directly used together with color histogram.
Currently, a widely used form of target representation is the color histogram,
which could be viewed as the discrete probability density function (PDF) of the
target region. Color histogram is an estimating mode of point sample distribution
and is very robust in representing the object appearance. However, using only color
histograms in mean shift tracking has some problems. First, the spatial information
of the target is lost. Second, when the target has similar appearance to the
background, color histogram will become invalid to distinguish them. For a better
target representation, the gradient or edge features have been used in combination
with color histogram. Several object representations that exploit the spatial
information have been developed by partitioning the tracking region into fixed size
fragments, meaningful patches or the articulations of human objects. For each
subregion, a color or edge feature based target model was presented.

The local binary pattern (LBP)16,17 technique is very effective to describe the
image texture features. LBP has advantages such as fast computation and rotation
invariance, which facilitates the wide usage in the fields of texture analysis,
image retrieval, face recognition, image segmentation, etc. Recently, LBP was successfully applied to the detection of moving objects via background
subtraction. In LBP, each pixel is assigned a texture value, which can be naturally
combined with the color value of the pixel to represent targets. In Ref. , Nguyen
et al. employed the image intensity and the LBP feature to construct a twodimensional
histogram representation of the target for tracking thermographic and
monochromatic video.

In this paper, we adopt the LBP scheme to represent the target texture feature
and then propose a joint color-texture histogram method for a more distinctive and
effective target representation. The major uniform LBP patterns are used to identify
the key points in the target region and then form a mask for joint color-texture
feature selection. The proposed target representation scheme eliminates smooth
background and reduces noise in the tracking process. Compared with the traditional
RGB color space based target representation, it efficiently exploits the target
structural information and hence achieves better tracking performance with fewer
mean shift iterations and higher robustness to various interferences of background
and noise in complex scenes.

The paper is organized as follows. Section 2 briefly introduces the mean shift
algorithm. Section 3 analyzes LBP and presents the joint color-texture histogram
scheme in detail. Experimental results are presented and discussed in Sec. 4.
Section 5 concludes the paper.

Download Here.