Hybrid Background Subtraction in Video
using Bi-level CodeBook Model
Soumya Varma
Federal Institute of Science and Technology,
Mahatma Gandhi University, Kerala
India
summu121@gmail.com
Sreeraj M
Federal Institute of Science and Technology,
Mahatma Gandhi University, Kerala
India
sreerajtkzy@gmail.com
Abstract—Detection of Objects in Video is a highly
demanding area of research. The Background Subtraction
Algorithms can yield better results in Foreground Object
Detection. This work presents a Hybrid CodeBook based
Background Subtraction to extract the foreground ROI from the
background. Codebooks are used to store compressed
information by demanding lesser memory usage and high speedy
processing. This Hybrid method which uses Block-Based and
Pixel-Based Codebooks provide efficient detection results; the
high speed processing capability of block based background
subtraction as well as high Precision Rate of pixel based
background subtraction are exploited to yield an efficient
Background Subtraction System. The Block stage produces a
coarse foreground area, which is then refined by the Pixel stage.
The system’s performance is evaluated with different block sizes
and with different block descriptors like 2D-DCT, FFT etc. The
Experimental analysis based on statistical measurements yields
precision, recall, similarity and F measure of the hybrid system
as 88.74%, 91.09%, 81.66% and 89.90% respectively, and thus
proves the efficiency of the novel system.
Keywords— Background Subtraction,Codebook Model,Foreground
Detection DCT, FFT
I. INTRODUCTION
Object Detection in Video has enormous applications in the
field of Target Recognition, Security Surveillance, Intelligent
Monitoring, Pedestrian Detection, Object Tracking etc. The
identification of an object in a video could be made very easily,
if the redundant video background is eliminated. In Computer
Vision related applications, Background Subtraction or
Filtering is a wide research area which focuses on subtracting
the unessential “background area” from the “foreground
region”. Any background subtraction methodology must not
demand much processing time and memory usage. Codebooks
are used to represent compressed form of information without
demanding much processing time and memory usage.
The remaining of the work is organized as follows. Section
II describes the Literature Review; Section III elucidates the
System Architecture. In Section IV, the details of the
Implementation are explained. Results and Evaluations are
described in Section V, and Conclusion and Future Scope of
the work are discussed in Section VI.
II. LITERATURE REVIEW
The process of Background Filtering is carried out by
analyzing the video frame-by-frame, maintaining a temporal
continuity between the consecutive frames of a video. The
approaches for Background Subtraction on the basis of
processing each frame could be broadly classified as pixel-
based and block-based methods. In the former approach, the
pixels of a frame that provide detailed information is taken
care of and decision is made on each pixel-whether it belongs
to foreground or background. But in the latter approach, each
frame segmented to different blocks of fixed size, is taken into
account and a decision is adopted for a block so that it is either
classified as foreground or background.
A. Pixel based methods
1) Frame Differencing
The simplest method of Background Subtraction is
the Frame Differencing strategy [1], in which the pixel
characteristics of a frame are subtracted from its previous
frame. In this approach, a pixel is classified as foreground if
the absolute difference between the pixel values in two
successive frames is greater than a predefined threshold. The
value for threshold must be carefully chosen to achieve
accurate results. However, this method offers only coarse form
of foreground which is least precise.
For any two frames, frame
i
and frame
i+1
, pixel P at
position (x, y), is classified as foreground or background as
follows
P
(x,y)=
P
fg
if |frame
i
- frame
i+1
|>Th (1)
P
bg
otherwise
where Th is the predefined threshold for classification. Here
the value that the parameter Th is highly sensitive in obtaining
the accurate output.
2) Background Modeling
A background subtraction strategy compares an
observed image with a background image. This process,
978-1-4799-2259-14/$31.00©2014
124