International Journal of Application or Innovation in Engineering & Management (IJAIEM)

Web Site: www.ijaiem.org Email: [email protected]

Volume 7, Issue 8, August 2018 ISSN 2319 - 4847

Design high- speed search by classifying binary

code book based on edge information in block
Maha A. Hameed
College of Science, Dept. of Astronomy, University of Baghdad

Abstract
While the encoding of a VQ based image coding requires a full codebook search for each input vector to find the best- matched
code word (i.e. it is a time consuming process), therefore, in this paper, we have adopted a new block- scanning technique for
generating the required sub code books by partition the code book depending on classified the code words and input image
blocks according to edge in blocks into empty, horizontal, vertical and diagonal categories. This classification is always
applying before the process that deals with code book. In our presence work, the classification idea is particularly simple for
software implementation where it could reduce the search time and the computational complexity, where it produce the same
decoded image quality as this obtain in the full search algorithm.
Keywords: Classification, binary code book, edge in blocks.

1. INTRODUCTION
The Block Truncation coding method preserves spatial details in the image content with low difficulty computation but
it has a medium compression ratio [1,2]. Therefore, in this paper, an adaptation ( vector quantization method) is
required to improve the compressibility of the BTC to represent the binary form of the coded image. The applications of
VQ method is limited because the encoding procedure is computationally intensive system [3,4]. Therefore, in this
paper, we present a new fast search algorithm for image compression using classified binary code book which will be
used to perform the encoding and decoding phases to improve the search of the binary code book. In this paper, to
decreasing the time of matching searching, consequently, simplify the computational complexity one can propose an
efficient image blocks classification and a new different categories binary code book ( sub code book i.e. empty,
horizontal, vertical and diagonal categories) by designing an algorithm for classified VQ based on edge in blocks,
where the classification idea is particularly simple for software implementation where it could reduce the search time
with same decoded image quality which may be obtained using full search.

2. The Block Truncation Compression technique

The BTC method was first proposed by Professors Mitchell and Delp [1,5]. This method can describe in general a
"FxF" pixels image then it is divided into blocks of "nxn" pixels, for each block, the mean and the standard deviation
are computed, where these values is difference from block to another, then a two level quantization for each block is
made as follows; If a pixel value is greater than or equal to the mean value it is assigned the value "1", otherwise "0".
Reconstruction is made with two values the mean and the standard deviation.

3. Vector Quantization (VQ)

Vector Quantization has been found to be a popular scheme in image compression where it can be processes in three
steps. There are designing code book, encoding vector, and decoding vector. In other word, this method can be
described as a code book is generated CB, CW 1 , CW 2 ,...., CWK , where K is the number of code words which stored in
the code book then an input vector (xi ) is then individually quantized to the closest code word in the code book by
searching in the code book to choose code word with minimal distortion by calculating the Squared Euclidean Distance
between the input vector and each code word in the code book [4]. The minimum square distortion for the vector xi is
found:

d i  d min( xi ,C wj ) .......... ..(1)

Compression is achieved by using the indexes of code words for the purpose of transmission and storage.
Reconstruction of the image can be implemented by technique called table look up, where the indexes are simply used
as addresses to the corresponding code words in the codebook [4,6].

4. Block classification
To decrease the computational difficulty of VQ method, we have adopted a new block- scanning technique for
generating the required code block. The proposed scanning method is depended on classifying the code book and input
Volume 7, Issue 8, August 2018 Page 25
International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: [email protected]
Volume 7, Issue 8, August 2018 ISSN 2319 - 4847

image blocks into empty, horizontal, vertical and diagonal categories. For each block, the direction of the major edge is
computed by finding the totality gradient value in the vertical and horizontal directions, as follows;

Let p(i,j) represents a nxn element in block of size nxn.

The vertical gradient is;
V(i , j )  p(i , j 1)  p (i , j ) .................(2) , where i=1…..n, j=1…..n-1.
While, the horizontal gradient is;
H (i , j )  p (i 1, j )  p (i , j ) ...............(3) , where i=1…..n-1, j=1…..n.
For each processed block, the effects of total vertical edge are computed as;
n n 1
EV   V(i , j ) ..................(4)
i 1 j 1
n n 1
E H   H (i , j ) .................(5)
j 1 i 1
Depending on equations 4 and 5 the classification procedure can be calculate as follows;

  TAN 1 ( EV ).............(6)
EH
Each block can be classified as empty when EV and EH are equal to zero (i.e.   0 ), but it is classified as horizontal
when 0    30 ,or it is classified vertical class when 60    90 , otherwise it is diagonal.
To overcome the problem connected with the code book search, In this paper, we use block classification (i.e. multi
code book design) based on the edge involved in block, this idea lead to reduce the searching time which required to
find the best matching between input image vector and the vectors in code book.

5. The procedure of fast algorithm

The proposed algorithm (i.e. block classification and multi code book designing based on the edge) is described as
follows;

Step1. Initialization, a binary code book blocks are classified to involve four sub code book ( i.e. empty, horizontal,
vertical and diagonal categories). Put a label for each class, "0" for empty, "1" for horizontal, "2" for vertical and "3"
for diagonal.
Step2. An image is divided into 4x4 blocks, then the BTC is implemented and for each block the reconstruction values
and bit plan ( binary block) are determined.
Step3. Take input block from binary image then calculate EV , EH and  for its.
Step4. From eq. 6 classify input block as empty when EV and EH are equal to zero, but it is classified as horizontal when
0    30 , or it is classified vertical class when 60    90 , otherwise it is diagonal class and give the input
block a label ( index "ic") of sub class ( i.e. 0,1,2,3).
Step5. Calculate the distortion of matching the input vector x with code words of the corresponding class ( i.e. sub code
book) then give the input block a label ( index "ib") of nearest code word.
Step6. Translate two reconstruction values with two indexes "ic" and "ib", then go to step 3.

Fig. 1: The diagram of the mentioned algorithm

Volume 7, Issue 8, August 2018 Page 26

International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: [email protected]
Volume 7, Issue 8, August 2018 ISSN 2319 - 4847

This method will be very efficient when we classify a code book, this idea lead to reducing in the searching time
which required to find the best matching and simplified a full code book search for each input vector to find the best-
matched code word. This is because many distortion computations can therefore be eliminated.

6. Experimental
We describe a new way to reducing VQ computation at the encoding and decoding by applying block classification to
present a new high speed search algorithm in VQ for improved BTC method which is examined by simulation. The
search in general code book has high complexity in computational, therefore the new adopted scanning method is used
for category code book blocks into sub smaller codebook then this leads to decreasing the complexity in computational
and more less time in searching than full search method.
At first a binary code book containing 256 code words is generated using the well- known LBG algorithm then the
code book is classified to involve four sub code book ( i.e. empty, horizontal, vertical and diagonal categories). The
image is divided into 4x4 blocks, the BTC is implemented, where the reconstruction values for each block and bit plan
are determined.

7. Conclusion
The effectiveness of applying classified method in the encoding and decoding algorithms are tested, where, the
elimination efficiencies of applying classified is listed in table 1.

Table 1: Elimination efficiencies for encoding and decoding of different images.

Image name Elimination efficiencies for encoding Elimination efficiencies for decoding
GIRL IMAGE 82.8% 58.2%
RMB1 IMAGE 67.6% 41.2%

From This table, one can see that the proposed method was very efficient when we classified a code book depending
on edge of its blocks into different categories, this idea was particularly simple for software implementation where it
leaded to reducing in the searching time which required to find the best matching then simplified a full code book
search for each input vector to find out the best- matched code word. In other word, the elimination efficiencies for
encoding and decoding of different images were very good this is because many distortion computations can therefore
be eliminated when classifying a code book.

References
[1] J. - M. Guo, " Improved block truncation coding using modified error diffusion", Electronics letters, Vol. 44, No. 7,
2008.
[2] Y. - Gi Wu, " Block truncation image bit plane coding", Optical Engineering, Vol. 41, No. 10, pp. 2476-2478,
2002.
[3] Sindhu M1 and Rajkamal R2, " Images and Its Compression Techniques – A Review", International Journal of
Recent Trends in Engineering, Vol. 2, No. 4, 2009.
[4] S. H. Hang and S.H. Chen, " Fast encoding algorithm for VQ based image coding", Electronics letters, No.19 , Vol.
26, pp. 1618-1619,1990.
[5] A. S. Devi1, S. S. Priyadharsini2 and S. Athinarayanan3, " A BLOCK BASED SCHEME FOR ENHANCING
LOW LUMINATED IMAGES", The International journal of Multimedia & Its Applications (IJMA), Vol.2, No.3,
2010.
[6] N. M. Nasrabadi and Y. Feng, " Vector quantization of images based upon a neural-network clustering algorithm",
Visual Communications and Image Processing, SPIE Vol. 1001, pp. 207-213, 1988.

AUTHOR
Maha Ahmed Hameed received the B.S. degree in Physics, University of Baghdad, College of Science, Dept. of
Physics and M.S. degree in Image Processing in University of Baghdad, College of Science, Dept. of Astronomy in
1999. She is teaching in the Astronomy Dept. now.

Volume 7, Issue 8, August 2018 Page 27