IRJET- Comparison and Simulation based Analysis of an Optimized Block Matching Algorithm with Four Step Search Algorithm

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 11 | Nov 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 1254
COMPARISON AND SIMULATION BASED ANALYSIS OF AN OPTIMIZED
BLOCK MATCHING ALGORITHM WITH FOUR STEP SEARCH ALGORITHM
Akanksha1, Jagbir Dhillon2, Rajbir Yadav3
1Akanksha, Assistant Professor, Department of ECE, NIET, Greater Noida
2Jagbir Dhillon, Principal, Department of ECE, Krishna Polytechnic College, Haryana, India
3Rajbir Yadav, Chairman, Department of ECE, Krishna Polytechnic College, Haryana, India
---------------------------------------------------------------------------***---------------------------------------------------------------------------
Abstract: With advancement in technology and increased
video applications, block based motion estimation is the
basic need for video coding systems. The four step search
block matching algorithm is very simple, robust and easy to
implement. It provides centre biased searching scheme and
halfway provision to stop in order to reduce the
computational cost. This algorithm checks for minimum of
17 points to find best match for each macro block with
minimum cost and has maximum of 27 checking points. It is
comparatively less complex than three step search. Thus, it is
widely used for various moving image applications such as
T. V monitoring, video conferencing etc. In this paper, a new
block matching algorithm is proposed which is very simple
and efficient. The proposed algorithm is also a four step
search with reduction in number of search points at every
step. Experimental results shows that proposed algorithm
achieves high PSNR using less number of search points and
requires less time for computations resulting in less
computational complexity.
Key Words: 4SS, Block matching algorithm, MAD,
Motion compensation, Motion estimation, MSE, MPC,
PSNR, SAD, Video compression
1. Introduction
Raw video consist of continuous sequence of frames or
moving images taken at various intervals of time. So, the
principles of video compression are based on image
compression. Advance in technology and new generations
bring an environment where there is high demand for
communication with moving picture or video. Now a day’s
video signal processing becomes an attractive area of
research. Digital videos are used in various applications
such as multimedia broadcasting, remote sensing, remote
monitoring, medical imaging, military etc and it is expected
that in future, real time video communication will be
common. Digital video is advantageous as compared to
analog video as it provides the interactive interface to
users. Digital videos are less prone to noise and easy to
transmit over medium. At present High definition videos
are more preferred due to its improved quality. A major
problem in video is that it requires large storage space as
well as large bandwidth for transmission.
However, there is limited channel bandwidth. So, for
efficient communication, the concept of video compression
is introduced. Video compression is a technique for
transforming video signal in compact representation
keeping the quality of the video signal as close as to the
original one. Compression is achieved by reducing the data
redundancy which occurs due to spatial, temporal, and
statistical co-relation between successive frames. The
MPEG format consists of three parts to reduce the data
redundancy. Motion estimation and compensation reduces
temporal redundancy between two frames in time domain.
The transform coding remove the spatial dependency in
frame and entropy coding is used to reduce statistical
redundancy over the residue. Motion estimation and
compensation is most important module on typical video
encoder. There are two approaches for motion estimation:
(1) pixel-based motion estimation, (2) block-based motion
estimation. Block based motion estimation is widely used
for video encoding. Motion estimation finds the
displacement between the two successive frames. One is
called current frame and the other is called as reference
frame. The reference frame occurs before and after the
current frame, called backward and forward motion
estimation. There is a very little change in spatial
arrangements of objects between the two frames in digital
video. That’s why it is good to send it over the network and
to store the difference between two successive frames.
Motion compensation finds the prediction of future frames
and the difference of compensated image and current
image is encoded and decoded. The difference frame is
called residue and it contain less details than the actual
frame. Due to reduction in details, video compression is
achieved. On the decoder side, decoded difference image is
added to compensate image to accurately synthesize
current frame from the previous frame. Thus, compression
efficiency is improved.

2. Block matching Algorithm
Block matching is an approach which is adopted to fulfil
the purpose of motion estimation in video compression,
aim of which is to reduce the temporal redundancy in
between the video sequences.
Fig. 1. Basic idea of block matching technique [Source:
Hussain Ahmed Choudhury et al., 2011]
Figure 1 below illustrates a method of block-matching rule.
Block matching algorithm is very simple and the initial one
for motion estimation. In this algorithm, each of the frames
is divided into group of macro blocks, which consist of
both the chrominance and luminance block. The
chrominance represent the colour and the luminance
represents the intensity/brightness [Source: Razali Yaakob
et al., 2013]. For coding efficiency, the motion estimation is
performed on luminance block. Each of the luminance
blocks in present frame is matched against the candidate
block lies in the search area on reference frame as shown
in figure 1. The best matched candidate block is observed
and its motion vector is recorded. For finding the best
match we compare the luminance block against each of the
candidates block which lies within the defined search area
that is constrained up to p pixels around the current block
in reference frame in both horizontal and vertical direction
[Neha Singh et al., 2016].
2.1 Existing Four-Step search algorithm
The four step search algorithm provides centre biased
searching scheme and halfway provision to stop in order
to reduce the computational cost. This algorithm was
developed in 1996 [S Sangeeta Mishra et al., 2014]. This
algorithm checks for minimum of 17 points to find best
match for each macro block with minimum cost and has
maximum of 27 checking points [S Immanuel Alex Pandian
et al., 2011].
Fig. 2. Four step search procedure [Source: Aroha
Barjatya, 2004]
3. Matching Criterion
As we know, block matching is a part of image matching. In
most of the image processing tasks, it is necessary to
examine two portions of image or two images on pixel to
pixel basis. These two images regions/images are chosen
from a spatial image sequence. The main aim of the
examination is to find the similarity in between these two
images or two image portions.
The similarity or correlation measure is main element
in matching process. On the other side, a more efficient
way to block matching is to find out minimum dissimilarity
or matching error. There are various types of matching
criterion, described below.
1. Mean of square error(MSE)
2. Sum of absolute difference(SAD)
3. Mean absolute difference(MAD)
4. Matching pixel count
For implementing the block motion estimation,
the current video frame is partitioned into various non-
overlapping blocks and the motion vector is calculated for
each of the candidate block with respect to reference
frame. For each of these matching criterions, a squared
block of size N×N symbols is considered. Let k is the
candidate frame and k-1 is past frame. Then, the intensity
of pixel co-ordinate (n1, n2) in frame k is given by S (n1, n2,
k).

1. Mean square error: Consider k-1 as past frame
then, for backward motion estimation, mean
square error of block of pixels computed at
displacement (i, j) in reference frame is given
using equation 1.
MSE (i, j) = ∑ ∑ ( )
( ) ……………. (eq. 1)
2. Sum of absolute difference: Like MSE, SAD
makes the error values positive but instead of
summing up the squared difference, the absolute
differences are summed up. SAD measure at
displacement (i, j) is defined using equation 2.
SAD (i, j) = ∑ ∑ ( )
( ) ..................... (eq. 2)
3. Mean absolute difference: In this criterion,
square of absolute differences is summed up and
is given using eq. 2.3.
MAD (i, j) = ∑ ∑ ( )
( ) ……....... (eq. 3)
4. Proposed Algorithm
The proposed algorithm is very simple and efficient. In
proposed algorithm, each step has two phases. The
search area is separated into 4 quadrants as shown in
figure 3(a). In first phase, the algorithm checks for
three locations A, B and C. A is at the origin and B and
C are at S = 4 locations in orthogonal directions away
from A as defined in figure 3(b).
Fig. 3. (a) Partition of search area into quadrants (b)
Search points in first phase of proposed algorithm [Source:
own elaboration]
Fig 4.3: Proposed block matching procedure [Source: own
elaboration]
Depending on the MAD calculated at 3 locations, the first
phase chooses one of the 4 possible quadrants to search
for second phase. The second phase finds the location with
the lowest weight and sets it as the new origin. Fig. 4
defines the complete proposed algorithm process started
with searching from centre location keeping s=4 and ends
at step size reduced to 0.5.
4. Simulation Results
We realised a video sequence Xylophone.mpg with
frames= 140. We consider the first 10 frames and value of
search parameter p is taken as 9 and find out the following
results in three different cases of macroblock.
 Case1: When Macro Block size = 4, P=9, No. of
frames= 10, Video sequence: Xylophone.mpg
Fig. 4. Computational complexity between 4SS and
proposed algorithm, mb=4

Figure 4 shows the graphs of computational complexity for
both the existing 4 step search and proposed algorithm
when the macroblock size is 4,number of frames=10 and
search parameter is taken as 9. In graph, the red line
shows the results for proposed algorithm and blue line
shows the results for existing algorithm. The graphs shows
that computational complexity of proposed algorithm is
less than the existing algorithm which means less number
of computations are required in case of new algorithm.
Fig. 5. Computation time required in 4SS and proposed
algorithm, Mb=4
Figure 5 shows the graphical results of computational time
required in both the existing 4 step search and proposed
algorithm when macro block size=4, p=9 and number of
frames=10. The graph curves and legend describe that less
computation time is required in case of proposed
algorithm than the existing one.
The similarities in structure of both the existing and
proposed algorithm when macroblock size taken as 4 is
described by the curves in figure 6. Their structures are
almost similar however, for proposed algorithm it is
somewhat small as it requires less number of search
points.
Fig. 6. Structural similarity between 4SS and existing
algorithm, Mb=4
Fig. 7. PSNR value for 4SS and existing algorithm, Mb=4
Figure 7 defines the two curves: red and blue. The red
curve shows graphical results for PSNR value in case of
proposed algorithm and the blue curve defines the
graphical results for PSNR value in case of existing 4 step
search algorithm. The results state that proposed
algorithm achieves high PSNR than existing algorithm
when macroblock size is taken as 4 and search parameter
value as 9 for 10 frames. The PSNR values, mean square
error values for both algorithms compared when mb=4,
p=9 and number of frames= 10 is shown in figure 8. Fig 8
also shows the step size taken at every step of search in
proposed algorithm.

Fig. 8. Results for MSE and PSNR values, mb=4, p=9 and
frames=10
 Case 2: When Macro Block size = 8, P=7, No. of
frames= 10, Video sequence: Xylophone.mpg.
Figure 9 shows the graphs of computational
complexity for both the existing 4 step search and
proposed algorithm when the macroblock size is
8,number of frames=10 and search parameter is
taken as 9. In graph, the red line shows the results for
proposed algorithm and blue line shows the results
for existing algorithm. The graphs shows that
computational complexity of proposed algorithm is
less than the existing algorithm which means less
number of computations are required in case of new
algorithm.
Fig. 9. Computational complexity between 4SS and
proposed algorithm, mb=8
In figure 10, the graphical results of computational time
required in both the existing 4 step search and proposed
algorithm when macro block size=8, p=9 and number of
frames=10 are shown. The graph curves and legend
describe that less computation time is required in case of
proposed algorithm than the existing one.
Fig. 10. Computation time required in 4SS and proposed
algorithm, Mb=8
The similarities in structure of both the existing and
proposed algorithm when macroblock size taken as 8 is
described by the curves in figure 11. Their structures are
almost similar however, for proposed algorithm it is
somewhat small as it requires less number of search
points.
Fig 11: Structural similarity between 4SS and existing
algorithm, Mb=8

Fig 12: PSNR value for 4SS and existing algorithm, Mb=8
Figure 12 defines the two curves: red and blue. The red
curve shows graphical results for PSNR value in case of
proposed algorithm and the blue curve defines the
graphical results for PSNR value in case of existing 4 step
search algorithm. The results state that proposed
algorithm achieves high PSNR than existing algorithm
when macroblock size is taken as 8 and search parameter
value as 9 for 10 frames. The PSNR values, mean square
error values for both algorithms compared when mb=8,
p=9 and number of frames= 10 is shown in figure 13. Fig
13 also shows the step size taken at every step of search in
proposed algorithm.
Fig 13: Results for MSE and PSNR values, mb=8, p=9 and
frames=10
5. Comparison Table
Table 1: Comparison between existing four step search
block matching algorithm and new proposed algorithm.
Video sequence: Xylophone.mpg
Existing 4 step search algorithm
Macro
block
size
Search
parameter
No. of
frames
SSIM MSE PSNR
4 9 10 0.944 18.205 38.4750
8 9 10 0.937 22.041 37.7116
New proposed algorithm
Macro
block
size
Search
parameter
No. of
frames
SSIM MSE PSNR
4 9 10 0.968 12.190 42.6296
8 9 10 0.964 15.045 41.8758
Table 1 show that on using same video sequence, no. of
frames, macroblock size and search parameter, the
proposed algorithm gives better results than the existing
algorithm. Like, when mb size is 4, p=9, no. of frames= 10,
the existing algorithm achieves SSIM=0.9445, MSE=
18.2051 and PSNR= 38.4750 while proposed algorithm
achieves SSIM=0.9684, MSE= 12.1908 and PSNR= 42.6296.
This shows that when macro block size decreases, PSNR
value increases and MSE decreases.
6. Conclusion
In this work, a new block matching algorithm has been
proposed for motion estimation process in video
compression that uses its neighbouring blocks for
predicting the motion vector of candidate block in current
frame. The existing 4ss search method is also developed
and compared. The results of proposed algorithm are
carried out on same video sequence for different macro
block size and its performance is compared with the
existing 4 step search. After the exhaustive study, it is
observed that the proposed algorithm outperforms the
existing search algorithms in terms of PSNR, computation
complexity and time. The proposed method uses fewer
numbers of blocks in search prediction thus, it requires
less computational time.
7. References
[1] Aroha Barjatya, “Block Matching Algorithms for Motion
Estimation”, DIP, 6620, Spring, 2004.
[2] Stefano Mattoccia, Federico Tombari, Luigi Di Stefano,
and Marco Pignoloni, “Efficient and optimal block
matching for motion estimation”, International Conference
on Image Analysis and Processing, IEEE, 2007.
[3] Najib Ben Aoun, Maher EL’ARBI, and Chokri Ben Amar,
“Multi resolution Motion Estimation and Compensation for
Video Coding”, IEEE Conference on Signal processing, pp.
1121-1124, 2010.
[4] S. Immanuel Alex Pandian, G. Josemin Bala, and Becky
Alma George, “A Study on Block Matching Algorithms for
Motion Estimation”, International journal on Computer

Science and Engineering, Vol.3, No. 1, pp. 34-44, January
2011.
[5] Hussain Ahmed Choudhury and Monjul Saikia, “Survey
on Block Matching Algorithms for Motion Estimation,
International Conference on Communication and Signal
Processing, IEEE, pp. 36-40, April 2011.
[6] Razali Yaakob, Alihossein Aryanfar, Alfian Abdul Halin
and Nasir Sulaiman, “A Comparison of Different Block
Matching Algorithms for Motion Estimation”, International
Conference on Electrical Engineering and Informatics,
Procedia Technology, pp. 199-205, 2013.
[7] S Sangeeta Mishra, Chittaranjan Pradhan, and Alka
Singh, “Comparative Study of Motion Estimation
Techniques in Video”, International Journal of Computer
Science and Information Technologies (IJCSIT), ISSN:
2982-2989, Vol. 5 (3), 2014
[8] Nasir Mahmud Khokhar and Wail Harasani,
“Performance Analysis of fast Block Matching Motion
Estimation Algorithms”, International Journal of Computer
Applications, Vol. 122, No. 6, pp. 15-20, July 2015.
[9] Neha Singh and Awanish Mishra, “Block Matching
Algorithm for Motion Estimation using Previous Motion
Vector Pattern”, International Journal of Computer
Applications, Vol. 150, No. 8, pp. 1-5, September 2016.
[10] Shailesh D. Kamble, Nileshsingh V. Thakur, and Preeti
R. Bajaj, “Modified Three-Step Search Block Matching
Motion Estimation and Weighted Finite Automata based
Fractal Video Compression”, International Journal of
Interactive Multimedia and Artificial Intelligence, Vol. 4,
No. 4, pp. 27-39, 2017.
Bibliography:
Akanksha has successfully completed her M.Tech in ECE
from BPSMV, Khanpur Kalan, Sonepat. Now, she is
currently working as an Assistant Professor in ECE
department and also as a trainer in IOT lab in NIET
College, Greater Noida.
Prof. Jagbir Dhillon is working as a Principal in Krishna
polytechnic College, Haryana. He is having long experience
of more than 20 years in the field of communication and
CATV network and he has specialisation in wireless
communication.
Prof. Rajbir Yadav is working as is Chairman in Krishna
polytechnic College, Haryana. He is having long experience
of more than 25 years in the field of education and he has
also attended many seminars on industry and institute
ineraction.

IRJET- Comparison and Simulation based Analysis of an Optimized Block Matching Algorithm with Four Step Search Algorithm

More Related Content

What's hot (19)

Similar to IRJET- Comparison and Simulation based Analysis of an Optimized Block Matching Algorithm with Four Step Search Algorithm (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Comparison and Simulation based Analysis of an Optimized Block Matching Algorithm with Four Step Search Algorithm