3.[18 30]graph cut based local binary patterns for content based image retrieval

Information and Knowledge Management www.iiste.org
ISSN 2224-5758 (Paper) ISSN 2224-896X (Online)
Vol 1, No.1, 2011

Graph Cut Based Local Binary Patterns for Content Based Image
Retrieval
Dilkeshwar Pandey
Department of Mathematics
Deen Bandhu Chotu Ram University of Science & Tech.
Murthal, Harayana, India
Email:dilkeshwar@hotmail.com

Rajive Kumar
Department of Mathematics, Deen Bandhu Chotu Ram University of Science & Tech.
Murthal, Harayana, India
E-mail: 2rajiv_kansal@yahoo.com
Abstract
In this paper, a new algorithm which is based on the graph cut theory and local binary patterns (LBP) for content
based image retrieval (CBIR) is proposed. In graph cut theory, each node is compared with the all other nodes
for edge map generation. The same concept is utilized at LBP calculation which is generating nine LBP patterns
from a given 3×3 pattern. Finally, nine LBP histograms are calculated which are used as a feature vector for
image retrieval. Two experiments have been carried out for proving the worth of our algorithm. It is further
mentioned that the database considered for experiments are Brodatz database (DB1), and MIT VisTex database
(DB2). The results after being investigated shows a significant improvement in terms of their evaluation
measures as compared to LBP and other existing transform domain techniques.
Keywords: Feature Extraction; Local Binary Patterns; Image Retrieval

1. Introduction
With the rapid expansion of worldwide network and advances in information technology there is an explosive
growth of multimedia databases and digital libraries. This demands an effective tool that allow users to search
and browse efficiently through such a large collections. In many areas of commerce, government, academia,
hospitals, entertainment, and crime preventions large collections of digital images are being created. Usually, the
only way of searching these collections was by using keyword indexing, or simply by browsing. However, as
the databases grew larger, people realized that the traditional keywords based methods to retrieve a particular
image in such a large collection are inefficient. To describe the images with keywords with a satisfying degree
of concreteness and detail, we need a very large and sophisticated keyword system containing typically several
hundreds of different keywords. One of the serious drawbacks of this approach is the need of trained personnel
not only to attach keywords to each image (which may take several minutes for one single image) but also to
retrieve images by selecting keywords, as we usually need to know all keywords to choose good ones. Further,
such a keyword based approach is mostly influenced by subjective decision about image content and also it is
very difficult to change a keyword based system afterwards. Therefore, new techniques are needed to overcome
these limitations. Digital image databases however, open the way to content based searching. It is common
phrase that an image speaks thousands of words. So instead of manual annotation by text based keywords,
images should be indexed by their own visual contents, such as color, texture and shape. The main advantage of
this method is its ability to support the visual queries. Hence researchers turned attention to content based image
retrieval (CBIR) methods. Several methods achieving effective feature extraction have been proposed in the
literature [Rui et al., Smeulders et al., kokare et al., and Liu et al.].
Swain et al. proposed the concept of color histogram in 1991 and also introduced the histogram intersection
distance metric to measure the distance between the histograms of images. Stricker et al. used the first three
central moments called mean, standard deviation and skewness of each color for image retrieval. Pass et al.
introduced color coherence vector (CCV). CCV partitions the each histogram bin into two types, i.e., coherent,
if it belongs to a large uniformly colored region or incoherent, if it does not. Huang et al. used a new color
feature called color correlogram which characterizes not only the color distributions of pixels, but also spatial
correlation of pair of colors. Lu et al. proposed color feature based on vector quantized (VQ) index histograms
in the discrete cosine transform (DCT) domain. They computed 12 histograms, four for each color component

18 | P a g e
www.iiste.org

Vol 1, No.1, 2011
from 12 DCT-VQ index sequences.
Texture is another salient and indispensable feature for CBIR. Smith et al. used the mean and variance of the
wavelet coefficients as texture features for CBIR. Moghaddam et al. proposed the Gabor wavelet correlogram
(GWC) for CBIR. Ahmadian et al. used the wavelet transform for texture classification. Moghaddam et al.
introduced new algorithm called wavelet correlogram (WC). Saadatmand et al. improved the performance of
WC algorithm by optimizing the quantization thresholds using genetic algorithm (GA). Birgale et al. and
Subrahmanyam et al. combined the color (color histogram) and texture (wavelet transform) features for CBIR.
Subrahmanyam et al. proposed correlogram algorithm for image retrieval using wavelets and rotated wavelets
(WC+RWC).
The recently proposed local binary pattern (LBP) features are designed for texture description. Ojala et al.
proposed the LBP and these LBPs are converted to rotational invariant for texture classification. Pietikainen et
al. proposed the rotational invariant texture classification using feature distributions. Ahonen et al. and Zhao et
al used the LBP operator facial expression analysis and recognition. Heikkila et al. proposed the background
modeling and detection by using LBP. Huang et al. proposed the extended LBP for shape localization. Heikkila
et al. used the LBP for interest region description. Li et al. used the combination of Gabor filter and LBP for
texture segmentation. Zhang et al. proposed the local derivative pattern for face recognition. They have
considered LBP as a nondirectional first order local pattern, which are the binary results of the first-order
derivative in images.
To improve the retrieval performance in terms of retrieval accuracy, in this paper, we proposed the graph cut
based local binary patterns (GCLBP) for CBIR. Two experiments have been carried out on Brodatz and MIT
VisTex databases for proving the worth of our algorithm. The results after being investigated show a significant
improvement in terms of their evaluation measures as compared to LBP and other existing transform domain
techniques.
The organization of the paper as follows: In section 1, a brief review of image retrieval and related work is
given. Section 2, presents a concise review of local binary patterns (LBP). Section 3, presents the feature
extraction, proposed system framework, and similarity measure. Experimental results and discussions are given
in section 4. Based on above work conclusions are derived in section 5.

2. 2. Local Binary Patterns
Ojala et al. proposed the local binary pattern (LBP) operator which describes the surroundings of a pixel by
generating a bit-code from the binary derivatives of a pixel as a complementary measure for local image
contrast. The LBP operator takes the eight neighboring pixels using the center gray value as a threshold. The
operator generates a binary code 1 if the neighbor is greater or equal than the center otherwise generates a binary
code 0. The eight neighboring binary code can be represented by a 8-bit number. The LBP operator outputs for
all the pixels in the image can be accumulated to form a histogram. Fig.1 shows an example of LBP operator.
For given a center pixel in the image, LBP value is computed by comparing it with those of its neighborhoods:
P −1
LBPP , R = ∑ 2i × f ( gi − g c )
i =0

(1)
1 x≥0
f ( x) = 
0 else
(2)
where g c is the gray value of the center pixel, g i is the gray value of its neighbors, P is the number of
neighbors and R is the radius of the neighborhood. Fig. 2 shows the examples of circular neighbor sets for
different configurations of ( P, R) .
The LBP measure the local structure by assigning unique identifiers, the binary number, to various micro-
structures in the image. Thus, LBP capture many structures in one unified framework. In the example in Fig.
3(b), the local structure is a vertical edge with a leftward intensity gradient. Other microstructures are assigned
different LBP codes, e.g., corners and spots, as illustrated in Fig. 4. By varying the radius R and the number of
samples P, the structures are measured at different scales, and LBP allows for measuring large scale structures
without smoothing effects, as is, e.g., the case for Gaussian-based filters.

19 | P a g e
www.iiste.org

Vol 1, No.1, 2011

Fig. 1: LBP calculation for 3×3 pattern

Fig. 2: Circular neighborhood sets for different (P,R)

Fig. 3. Illustration of LBP. (a) The LBP filter is defined by two parameters; the circle radius R and the number of
samples P on the circle. (b) Local structure is measured w.r.t. a given pixel by placing the center of the circle in
the position of that pixel. (c) Samples on the circle are binarized by thresholding with the intensity in the center
pixel as threshold value. Black is zero and white is one. The example image shown in (b) has an LBP code of
124. (d) Rotating the example image in (b) 900 clockwise reduces the LBP code to 31, which is the smallest
possible code for this binary pattern. This principle is used to achieve rotation invariance.

Fig. 4: Various microstructures measured by LBP. The gray circle indicates the center pixel. Black and white
circles are binarized samples; black is zero and white is one.
After identifying the LBP pattern of each pixel (j, k), the whole image is represented by building a histogram:

20 | P a g e
www.iiste.org

Vol 1, No.1, 2011
N1 N2
H LBP (l ) = ∑∑ f (LBP( j , k ), l ); l ∈ [0, (2 P − 1)]
j =1 k =1

(3)
1 x= y
f ( x, y ) = 
0 else
(4)
where the size of input image is N1 × N 2 .

3. 3. Feature Extraction
The weighted graph (Li Xi et al.,) with no self loops is G = (V , E ,W ) , where V = {1, 2,......., N } the node set is
(N=m.n is the total number of pixels in Q ∈ R m×n ) E ⊆ V × V represents the edge set, and W = ( wij ) denotes
N×N

an affinity matrix with the element wij being the edge weight between nodes i and j.
Based on the above graph cut theory we compare the each pixel of 3×3 pattern with remaining eight pixel gray
values for generating binary code. Finally, nine LBP patterns are collected for LBP histogram calculation and
these are used as a feature vector for image retrieval. The flowchart of the proposed system is shown in Fig. 5
and algorithm for the same is given below:

3.1 Proposed System Framework (GCLBP)
Algorithm:
Input: Image; Output: Retrieval Result
1. Load the input image.
2. Collect the 3×3 pattern for a center pixel i.
• Construct the graph cut for 3×3 pattern.
• Generate nine LBP patterns.
• Go to next center pixel.
3. Calculate the graph cut LBP (GCLBP) histograms.
4. Form the feature vector by concatenating the nine LBP features.
5. Calculate the best matches using Eq. (5).
6. Retrieve the number of top matches.

Fig. 5: Proposed system framework

3.2 Similarity Measurement
In the presented work d1 similarity distance metric is used as shown below:

21 | P a g e
www.iiste.org

Vol 1, No.1, 2011
Lg f I1 ,i − fQ ,i
D(Q, I1 ) = ∑
i =1 1 + f I1 ,i + f Q ,i
(5)
where Q is query image, Lg is feature vector length, I1 is image in database; f I ,i is ith feature of image I in the
database, f Q ,i is ith feature of query image Q.

4. Experimental Results and Discussions
For the work reported in this paper, retrieval tests are conducted on two different databases (Brodatz, and MIT
VisTex) and results are presented separately.

4.1. Database (DB1)
The database DB1 used in our experiment that consists of 116 different textures comprising of 109 textures
from Brodatz texture photographic album [Brodatz P.], seven textures from USC database
[http://sipi.usc.edu/database/]. The size of each texture is 512 × 512 and is further divided into sixteen 128 × 128
non-overlapping sub-images, thus creating a database of 1856 (116 × 16) images.
No. of Relevant Images Retrieved
Precision ( P) = × 100
Total No. of Images Retrieved
(6)
N1
1
Group Precision (GP ) = ∑P
N1 i =1
(7)
1 Γ1
Average Retrieval Precision ( ARR ) = ∑ GP
Γ1 j =1
(8)
Number of relevant images retrieved
Recall ( R) =
Total Number of relevant images
(9)
1 N1
Group Recall (GR) = ∑R
N1 i =1
(10)
Γ1
1
Average Retrieval Rate ( ARR) = ∑ GR
Γ1 j =1
(11)
where N1 is number of relevant images and Γ1 is number of groups.

Table 1: Retrieval results of proposed method (GCLBP) and LBP in terms of average retrieval precision (ARP)
(%)
Number of top matches considered
Method
1 3 5 7 9 11 13 15 16
LBP 100 89.17 84.67 81.71 79.01 76.33 73.86 71.18 69.65
GCLBP 100 93.19 89.73 87.27 85.02 82.71 80.47 77.88 76.45

Table 2: Retrieval results of proposed method (GCLBP) and LBP in terms of average retrieval rate (ARR) (%)
Method
16 32 48 64 80 96 112
LBP 69.65 80.16 84.47 87.05 89.02 90.44 91.63

22 | P a g e
www.iiste.org

Vol 1, No.1, 2011
GCLBP 76.45 84.57 87.85 89.79 91.13 92.18 93.03
DT-CWT 74.16 83.83 87.13 89.11 90.48 91.48 92.3
DT-RCWT 72.33 80.88 84.32 86.28 87.82 88.98 89.92

Table 3: Performance of proposed method (GCLBP) with different distance measures in terms of average
retrieval rate (ARR) (%)
Distance Number of top matches considered
Method
Measure 16 32 48 64 80 96 112
Manhattan 79.89 86.82 89.61 91.30 92.46 93.34 94.04
Canberra 77.73 85.09 88.24 90.02 91.37 92.32 93.07
GCLBP
Euclidean 78.81 85.59 88.43 90.24 91.54 92.47 93.25
d1 76.45 84.57 87.85 89.79 91.13 92.18 93.03

Fig. 6: comparison of proposed method (GCLBP) with LBP on DB1 database in terms of ARP

Table 1 and Fig. 6 summarize the retrieval results of the proposed method (GCLBP), and LBP in terms of
average retrieval precision and Table 2 and Fig. 7 illustrate the performance of proposed method (GCLBP), LBP
and other transform domain techniques in terms of average retrieval rate. Table 3 and Fig. 8 summarize the
performance of proposed method (GCLBP) with different distance measures in terms of average retrieval rate.

23 | P a g e
www.iiste.org

Vol 1, No.1, 2011

Fig. 7: Comparison of proposed method (GCLBP) with: (a) LBP on DB1 database in terms of ARR, (b) with
LBP and other transform domain features on DB1 database in terms of ARR.
From the Tables 1 to 3 and Fig. 6 to 8 the following can be observed:
1. The average retrieval precision of proposed method (GCLBP) (100% to 76.45%) is more as compared to
LBP (100% to 69.65%).
2. The average retrieval rate of GCLBP (76.45% to 93.03%) is more compared to LBP (69.65% to
91.63%), DT-CWT (74.16% to 92.3%), and DT-RCWT (72.33% to 89.92%).
3. The performance of the proposed method with Manhattan distance (79.89% to 94.04%) is more as
compared to Canberra (77.73% to 93.07%), Euclidean (78.81% to 93.25%), and d1 distance (76.45% to
93.03%).
From Tables 1 to 3, Fig. 6 to 8, and above observations, it is clear that the proposed method is outperforming the
LBP and other transform domain techniques. Fig. 9 illustrates the retrieval results of query image based on the
proposed method (GCLBP).

24 | P a g e
www.iiste.org

Vol 1, No.1, 2011

Fig. 8: Performance of proposed method (GCLBP) with different distance measures on DB1 database in terms
of ARR.

4.2. Database DB2
The database DB2 used in our experiment consists of 40 different textures
[http://vismod.www.media.mit.edu]. The size of each texture is 512 × 512 . Each 512 × 512 image is divided into
sixteen 128 ×128 non-overlapping sub-images, thus creating a database of 640 (40 × 16) images. The
performance of the proposed method is measured in terms of ARP and ARR.

Table 4: Retrieval results of proposed method (GCLBP) and LBP in terms of average retrieval precision (ARP)
(%)
Method
1 3 5 7 9 11 13 15 16
LBP 100 93.85 90.90 88.37 85.45 82.69 79.85 76.35 74.39
GCLBP 100 97.13 95.25 93.05 90.45 87.52 84.87 81.46 79.44

25 | P a g e
www.iiste.org

Vol 1, No.1, 2011

Fig. 9: Retrieval results of proposed method (GCLBP) of query image: (a) 1, (b) 724, and (c) 1850 of database
DB1.
Table 4 and Fig. 10 summarize the retrieval results of the proposed method (GCLBP) and LBP in terms of
average retrieval precision and Table 5 and Fig. 11 illustrate the performance of proposed method (GCLBP) and
LBP in terms of average retrieval rate. Table 6 and Fig. 12 summarize the performance of proposed method
(GCLBP) with different distance measures in terms of average retrieval rate.
From the Tables 4 to 6 and Fig. 10 to 12 the following can be observed:
1. The average retrieval precision of proposed method (GCLBP) (100% to 79.44%) is more as compared to

26 | P a g e
www.iiste.org

Vol 1, No.1, 2011
LBP (100% to 74.39%).
2. The average retrieval rate of GCLBP (79.44% to 97.24%) is more compared to LBP (74.39% to
97.08%).
3. The performance of the proposed method with d1 distance (79.44% to 97.24%) is more as compared to
Canberra (74.7% to 93.48%), Euclidean (80.07% to 97.20%), and Manhattan distance (80.47% to
95.46%).
From Tables 4 to 6, Fig. 10 to 12, and above observations, it is clear that the proposed method is outperforming
the LBP and other transform domain techniques.

Fig. 10: comparison of proposed method (GCLBP) with LBP on DB2 database in terms of ARP

Table 5: Retrieval results of proposed method (GCLBP) and LBP in terms of average retrieval rate (ARR) (%)
Method
16 32 48 64 80 96 112
LBP 74.39 86.69 91.14 93.77 95.35 96.36 97.08
GCLBP 79.44 88.36 92.00 94.22 95.54 96.53 97.24

27 | P a g e
www.iiste.org

Vol 1, No.1, 2011

Fig. 11: Comparison of proposed method (GCLBP) with LBP on DB2 database in terms of ARR.

Fig. 12: Performance of proposed method (GCLBP) with different distance measures on DB2 database in terms
of ARR.

Table 6: Performance of proposed method (GCLBP) with different distance measures in terms of average
retrieval rate (ARR) (%)
Distance Number of top matches considered
Method
Measure 16 32 48 64 80 96 112
Manhattan 80.47 88.62 91.560 93.21 94.23 94.97 95.46
Canberra 74.70 84.57 88.260 90.58 92.00 92.81 93.48
GCLBP
Euclidean 80.07 88.44 92.07 94.17 95.56 96.52 97.20
d1 79.44 88.36 92.00 94.22 95.54 96.53 97.24

28 | P a g e
www.iiste.org

Vol 1, No.1, 2011
5. Conclusion
A new algorithm which is based on the graph cut theory and local binary patterns (LBP) for content based image
retrieval (CBIR) is proposed in this paper. The proposed method extracts the nine LBP patterns from a given
3×3 pattern and these are used as the features. Two experiments have been carried out for proving the worth of
our algorithm. The results after being investigated shows a significant improvement in terms of their evaluation
measures as compared to LBP and other existing transform domain techniques.

References
Ahonen T., Hadid A., Pietikainen M., Face description with local binary patterns: Applications to face
recognition, IEEE Trans. Pattern Anal. Mach. Intell., 28 (12): 2037-2041, 2006.
Ahmadian A., Mostafa A. (2003), An Efficient Texture Classification Algorithm using Gabor wavelet, 25th
Annual international conf. of the IEEE EMBS, Cancun, Mexico, 930-933.
Birgale L., Kokare M., Doye D. (2006), Color and Texture Features for Content Based Image Retrieval,
International Conf. Computer Grafics, Image and Visualisation, Washington, DC, USA, 146 – 149.
Brodatz P. (1996), “Textures: A Photographic Album for Artists and Designers,” New York: Dover.
Heikkil M., Pietikainen M., A texture based method for modeling the background and detecting moving objects,
IEEE Trans. Pattern Anal. Mach. Intell., 28 (4): 657-662, 2006.
Heikkila M., Pietikainen M., Schmid C., Description of interest regions with local binary patterns, Elsevie J.
Pattern recognition, 42: 425-436, 2009.
Huang J., Kumar S. R., and Mitra M., Combining supervised learning with color correlograms for content-based
image retrieval, Proc. 5th ACM Multimedia Conf., (1997) 325–334.
Kokare M., Chatterji B. N., Biswas P. K., A survey on current content based image retrieval methods, IETE J.
Res., 48 (3&4) 261–271, 2002.
Liu Ying, Dengsheng Zhang, Guojun Lu, Wei-Ying Ma, Asurvey of content-based image retrieval with high-
level semantics, Elsevier J. Pattern Recognition, 40, 262-282, 2007.
Li M., Staunton R. C., Optimum Gabor filter design and local binary patterns for texture segmentation, Elsevie
J. Pattern recognition, 29: 664-672, 2008.
Li Xi, Hu Weiming, Zhang Zhongfei, and Wang Hanzi, Heat Kernel Based Local Binary Pattern for Face
Representation, IEEE Signal Processing Letters, 17 (3) 308–311 2010.
Lu Z. M. and Burkhardt H., Colour image retrieval based on DCT domain vector quantization index histograms,
J. Electron. Lett., 41 (17) (2005) 29–30.
MIT Vision and Modeling Group, Vision Texture. [Online]. Available: http://vismod.www.media.mit.edu.
Moghaddam H. A., Khajoie T. T., Rouhi A. H and Saadatmand T. M. (2005), Wavelet Correlogram: A new
approach for image indexing and retrieval, Elsevier J. Pattern Recognition, 38 2506-2518.
Moghaddam H. A. and Saadatmand T. M. (2006), Gabor wavelet Correlogram Algorithm for Image Indexing
and Retrieval, 18th Int. Conf. Pattern Recognition, K.N. Toosi Univ. of Technol., Tehran, Iran, 925-928.
Moghaddam H. A., Khajoie T. T. and Rouhi A. H. (2003), A New Algorithm for Image Indexing and Retrieval
Using Wavelet Correlogram, Int. Conf. Image Processing, K.N. Toosi Univ. of Technol., Tehran, Iran, 2 497-
500.
Ojala T., Pietikainen M., Harwood D., A comparative sudy of texture measures with classification based on
feature distributions, Elsevier J. Pattern Recognition, 29 (1): 51-59, 1996.
Ojala T., Pietikainen M., Maenpaa T., Multiresolution gray-scale and rotation invariant texture classification
with local binary patterns, IEEE Trans. Pattern Anal. Mach. Intell., 24 (7): 971-987, 2002.
Pass G., Zabih R., and Miller J., Comparing images using color coherence vectors, Proc. 4th ACM Multimedia
Conf., Boston, Massachusetts, US, (1997) 65–73.
Pietikainen M., T. Ojala, T. Scruggs, K. W. Bowyer, C. Jin, K. Hoffman, J. Marques, M. Jacsik, W. Worek,
Overview of the face recognition using feature distributions, Elsevier J. Pattern Recognition, 33 (1): 43-52,
2000.
Rui Y. and Huang T. S., Image retrieval: Current techniques, promising directions and open issues, J.. Vis.
Commun. Image Represent., 10 (1999) 39–62.

29 | P a g e
www.iiste.org

Vol 1, No.1, 2011
Saadatmand T. M. and Moghaddam H. A., Enhanced Wavelet Correlogram Methods for Image Indexing and
Retrieval, IEEE Int. Conf. Image Processing, K.N. Toosi Univ. of Technol., Tehran, Iran, (2005) 541-544.
Saadatmand T. M. and Moghaddam H. A., A Novel Evolutionary Approach for Optimizing Content Based
Image Retrieval, IEEE Trans. Systems, Man, and Cybernetics, 37 (1) (2007) 139-153.
Smeulders A. W.M., Worring M., Santini S., Gupta A., and Jain R., Content-based image retrieval at the end of
the early years, IEEE Trans. Pattern Anal. Mach. Intell., 22 (12) 1349–1380, 2000.
Smith J. R. and Chang S. F., Automated binary texture feature sets for image retrieval, Proc. IEEE Int. Conf.
Acoustics, Speech and Signal Processing, Columbia Univ., New York, (1996) 2239–2242.
Stricker M. and Oreng M., Similarity of color images, Proc. SPIE, Storage and Retrieval for Image and Video
Databases, (1995) 381–392.
Subrahmanyam M., Gonde A. B. and Maheshwari R. P., Color and Texture Features for Image Indexing and
Retrieval, IEEE Int. Advance Computing Conf., Patial, India, (2009) 1411-1416.
Subrahmanyam Murala, Maheshwari R. P., Balasubramanian R., A Correlogram Algorithm for Image Indexing
and Retrieval Using Wavelet and Rotated Wavelet Filters, Int. J. Signal and Imaging Systems Engineering.
Swain M. J. and Ballar D. H., Indexing via color histograms, Proc. 3rd Int. Conf. Computer Vision, Rochester
Univ., NY, (1991) 11–32.
Tan X. and Triggs B., Enhanced local texture feature sets for face recognition under difficult lighting conditions,
IEEE Tans. Image Proc., 19(6): 1635-1650, 2010.
University of Suthern California, Signal and Image Processing Institute, Rotated Textures. [Online]. Available:
http://sipi.usc.edu/database/.
Zhao G., Pietikainen M., Dynamic texture recognition using local binary patterns with an application to facial
expressions, IEEE Trans. Pattern Anal. Mach. Intell., 29 (6): 915-928, 2007.

30 | P a g e
www.iiste.org

3.[18 30]graph cut based local binary patterns for content based image retrieval

More Related Content

What's hot (20)

Viewers also liked (7)

Similar to 3.[18 30]graph cut based local binary patterns for content based image retrieval (20)

More from Alexander Decker (20)

Recently uploaded (20)

3.[18 30]graph cut based local binary patterns for content based image retrieval