A Modified Local Binary Pattern (LBP) for Content-based Image Retrieval
Mamta Martolia1,*, Nilesh Dhanore2, Anupam Singh3, Vivek Shahare4, Nitin Arora5
1
Uttarakhand Technical University, Dehradun, India
2
Jhulelal Institute of Technology, Nagpur, India
3,4
University of Petroleum & Energy Studies, Dehradun, India
5
Electronics & Computer Discipline, Indian Institute of Technology, Roorkee, India
Email:
1[email protected],
2[email protected],
3
[email protected],
4[email protected],
5
[email protected]
*Correspondence: [email protected]
Abstract
Researchers consider Content-Based Image Retrieval (CBIR) as one of the challenging ground as the searching of an image does not depend on manually assigned annotations. Instead, it uses discriminative features to retrieve a query image. Local Binary Pattern (LBP) technique is extensively used in literature to extract texture features; however, it takes more execution time as it considered all the 8 bits while calculating LBP values. In this paper, we proposed a modified version of the LBP (m- LBP) technique that uses only the 4 most significant bits (MSB). The robustness of the proposed method is investigated on two publicly available datasets, namely WANG and SIMPSON images. The similarity score is measured using Euclidean and Manhattan distance-based metrics. Experimental results reveal the effectiveness of the proposed algorithm in terms of precision and recall values and execution time in comparison to the existing state of the art techniques.
Keywords: Image processing, texture analysis, LBP, MSB, CBIR, Euclidean distance, Manhattan distance
1. Introduction
Image is an essential part of daily life. There are many images of data generated every day by using electronic gadgets. Manually searching for any particular image from a large set of images is not an easy task. Text-based image retrieval (TBIR) is a technique that is used to search an image in a database of the image using text annotation. But, TBIR is a very time taking and not an efficient technique because in the TBIR technique user has to check every image one by one manually. Sometimes different users have different perceptions about the images. For example, if one image containing greenery and animals, both then user can name that image based on greenery or based on the animal name. LBP technique is first proposed in the year 1994. LBP used the local content of any image.
There are three types of content of an image, namely color content, shape content, and texture content. LBP focuses on the texture content of an image. LBP technique considers the eight local neighbors (figure 1) of every pixel and calculates the difference between the intensity of the central pixel and neighborhood pixels and generates an eight bits binary pattern (eq. 1 and eq. 2). For the corner pixels, the intensity of only three pixels is
available. For the rest of the five pixels, padding is used, and we padded intensity with the zero values. This paper proposed a modified LBP technique (m-LBP), which uses only 4 most significant bits (MSB). The proposed method is applied to the data set of images, and the results are compared with the LBP technique.
1.1 Local Binary Pattern
Local binary pattern considers a 3x3 matrix of a pixel and its local neighborhood eight pixels. It calculates the 8 bits pattern based on the difference between the intensity of the middle pixel and its eight neighbors (eq.1) [10]. In this, if the difference between the central pixel and neighbor pixel is greater than 0, then the bit is set to 1 else the bit is set 0. The calculation of the LBP value of a pixel is explained in figure 1.
1 1
( , ) 2 ( , )
N i
i c
i
LBP N R D I I
(1)( , ) 1 ,
0 ,
i C
i c
if I I D I I
Otherwise
(2)
Where N is representing the number of neighboring pixels, and R is representing radius. Ic
and Ii are representing the intensity of the central pixel and ith neighbor pixel, respectively.
I3 I2 I1 39 56 30 0 1 0 4 2 1 0 2 0
I4 Ic I8 48 45 50 1 1 8 128 8 154 128
I5 I6 I7 49 24 26 1 0 0 16 32 64 16 0 0
(i) (ii) (iii) (iv) (v)
Fig.1. Diagrams (i) to (v) are presenting the calculation of Local Binary Pattern (LBP) value for a particular pixel. (i) a 3×3 window with universal
representations of the central and its eight neighboring pixels. (ii) an example of a 3x3 window with centered pixel’s intensity 45 and eight adjacent pixel’s intensities as shown (iii) pixels with centered and binary values 0 and 1 are assigned as a result based on sign of difference values (iv) specific weights binary digits 0 and 1 (v) multiplication with weights and
the addition of all the values to get LBP value (154).
2.
Related work
In this section, we discussed the work done by many researchers on different variants of LBP. The author described a novel approach for image retrieval. Xiaoyu Wang et al.
(2009) [1] proposed a unique human detection method by combining histograms of oriented gradients (HOG) and local binary pattern(LBP) as the feature set and achieved a detection rate of 91.3%. Zhenhua Guo et al. (2010) [2] proposed a hybrid scheme, globally rotation invariant matching with locally variant LBP texture features. The proposed LBPV operator and global matching scheme achieved significant improvement, sometimes more than 10% in terms of classification accuracy, over traditional locally rotation invariant LBP method. Lun Zhang et al. (2007) [3] proposed a concept of a multi- block local binary pattern (MB-LBP). The MB-LBP encodes rectangular regions‟
intensities by local binary pattern operator, and the resulting binary patterns can describe
diverse local structures of images. Sheryl Brahnam et al. (2014) [4] represents many LBP variants with applications. N. Arora et al. (2019) [5] presented local octal patterns and local hexadecimal pattern to retrieve the image. Juha Ylioinas et al. (2012) [6] proposed a method for age classification in unconstrained conditions using LBP. The experimental analysis points out the complexity of the age classification problem under uncontrolled settings. The proposed method provides state-of-the-art performance that can be used as a reference for future investigations. Loris Nanni et al. (2012) [7] presented a survey on LBP based texture descriptors for image classification. A Hadid et al. (2015) [8]
introduced a comparative analysis using 13 variants of local binary patterns. All these variants are compared, and based on this conclusion is given. A. K. Bedi et al. (2018) [9]
presented a review of LBP variants in the spatial domain. Timo Ahonen et al. (2004)[11]
presented face recognition technique using shape and texture features. For texture features, they have used the LBP technique. Turgay Celik et al. (2009) [12] presented multi-scale texture classification using the dual-tree complex wavelet transform. S.
Chakraborty et al. (2017) [13] proposed a local pattern descriptor in high order derivative space. The proposed descriptor significantly reduces the extraction as well as matching time during recognition. S. R. Dubey et al. (2016) [14] presented a novel image feature descriptor that is based on the local bit-plane decoded pattern for indexing and retrieval of biomedical images. Image retrieval using hybrid features and the neural network has been described in [15]. Izem Hamouchene et al. (2014) [16] proposed a new texture analysis approach for Iris recognition. They also introduced a method using mean and variance for solving rotation invariant problem. Y. He et al. (2013) [17] proposed a multi-structure local binary pattern for texture classification. A multi-structure local binary pattern (MS- LBP) operator is achieved by executing the extended LBP operator on different layers of an image pyramid. Kokare M. et al. (2006) [18] presented a technique for rotation- invariant texture image retrieval using rotated complex wavelet filters. Kokare M. et al.
(2007) [19] presented a technique texture image retrieval using rotated wavelet filters.
Nanni L et al. (2010) [20] focused on the use of image-based machine learning techniques in medical image analysis. They presented some variants of local binary patterns (LBP), which are widely considered state of the art among texture descriptors. N. Kazak et al.
(2018) [21] presented two spiral and four spiral LBP for classification and retrieval purposes. They have shown that spiral LBP achieved the highest accuracy among all the LBP variants. Zhenhua Guo et al. (2010) [22] presented rotation invariant texture classification using LBP variance (LBPV) with global matching. One of the improved techniques using row, column, and diagonal pixels has been described in [23].
2.1 Main contribution
Several local patterns have been developed for image retrieval until now. However, most of the existing local patterns (Shengcai Liao et al., 2007; Moore & Bowden, 2011; Murala et al., 2012b; Nanni, Lumini, & Brahnam, 2010; B. Zhang et al., 2010) compare the intensity of a center pixel in a 3×3 window (see Fig. 1) with one of its 8 neighboring pixels at a time to encode it in binary form. In this paper, we proposed a new texture technique, which is based on 4 most significant bits instead of 8 bits as in state-of-art LBP techniques. By considering only 4 most significant bits, results show that the proposed method also produced the same results as in the case of LBP in less executing time. The process of calculating LBP values using the proposed method is discussed in fig. 2.
2.2 Modified Local Binary Pattern
In the proposed technique, we consider the only four most significant bits. The modified LBP (eq. 2) consider only for local neighbors, and it generated a four bits pattern based on the difference between a central pixel and four neighbors. The calculation of modified LBP using the most significant bits is explained in figure 2.
4 5 6 7
5 6 7 8
( , ) 2 ( , c) 2 ( , c) 2 ( , c) 2 ( , c) MSBLBP N R D I I D I I D I I D I I (3)
( , ) 1 ,
0 ,
i C
i c
if I I D I I
Otherwise
(4)
Where N is representing the number of neighboring pixels (we assumed 8 in this paper), and R is representing radius. Ic and Ii are representing the intensity of the central pixel and ith neighbor pixel, respectively.
I3 I2 I1 39 56 30 0 1 0 4 2 1 0 2 0
I4 Ic I8 48 45 50 1 1 8 128 8 144 128
I5 I6 I7 49 24 26 1 0 0 16 32 64 16 0 0
(i) (ii) (iii) (iv) (v)
Fig.2. Diagrams (i) to (v) are presenting the calculation of Local Binary Pattern (LBP) value using only four most significant bits for a particular pixel. (i) a 3×3 window with universal
representations of the central and its eight neighboring pixels. (ii) an example of a 3x3 window with centered pixel’s intensity 45 and eight adjacent pixel’s intensities as shown (iii)
pixels with centered and binary values 0 and 1 are assigned as a result based on sign of difference values (iv) specific weights binary digits 0 and 1 (v) multiplication with weights
and the addition of all the values to get modified LBP value (144).
2.3 Advantage over others
In the modified LBP (m-LBP) technique, which is based on the 4 most significant bits (MSB) of an 8-bit binary pattern, the main advantage is less execution time. By using the proposed technique, the same results are produced within less time as compared with the state-of-art LBP and its variants. The proposed LBP technique is based only on the 4 most significant bits, and therefore the time of execution is almost half in this method.
3 Proposed System framework
Fig. 3. Proposed system framework 3.1 System framework algorithm
The proposed framework algorithm works in two phases. In phase 1 (step 1 to step 3), we calculated feature vectors and distance measure (eq. 7 and eq.8) using LBP. In phase 2 (step 4 to step 6), we computed feature vectors and distance measure (eq. 7 and eq.8)using the proposed LBP. At last, all the computed distances are compared in step 7.
Algorithm: Algorithm for system framework Input: Set of images and query image Output: Image similar to the query image Initialization Phase 1:
1. Calculate feature vector of query image using LBP texture descriptor
2. for i = 1 to n do // n is the number of images in the dataset
3. Calculate feature vector of all the image in the dataset using LBP texture descriptor 4. end for
5. for i= 1 to n do
6. Calculate Euclidean distance and Manhattan distance between dataset images feature
7. vectors and query image feature vector 8. end for
9. return Image with the smallest distance Initialization Phase 2:
10. Calculate feature vector of query image using the proposed m-LBP texture descriptor
11. for i = 1 to n do // n is the number of images in the dataset
12. Calculate feature vector of all the image in the dataset using m-LBP texture descriptor
13. end for 14. for i= 1 to n do
15. Calculate Euclidean distance and Manhattan distance between dataset images feature
16. vectors and query image feature vector 17. end for
18. return Image with the smallest distance 3.2 Similarity measure
Various similarity measure techniques like d1 distance, Euclidian distance, Manhattan distance, Canberra distance, Chi-square distance are existed for calculating the distance between the query image and the retrieved image. In this paper, we used the Euclidean distance and Manhattan distance measure technique.
a. Euclidian Distance
Euclidian distance is the linear distance between features vectors. In the case of two similar feature vectors, Euclidian distance is 0. Mathematically Euclidian distances between two feature vectors can be calculated with the help of eq. 7.
2 1/ 2 1
( | | )
n
dj qj
i
D F F
(7) Where D is the distance, and Fdj is a feature vector of the jth image in the database, and Fqj
is the jth query image.
b. Manhattan Distance
Manhattan distance is another way to measure the distance between two feature vectors.
Mathematically, Manhattan distances between two feature vectors can be calculated with the help of eq. 8.
1
( | |)
n
dj qj
i
D F F
(8)Where D is the distance, and Fdj is a feature vector of the jth image in the database, and Fqj
is the jth query image.
4 Experimental results and Analysis
Both the techniques have been tested on a 64bit operating system with Intel (R) Core (TM) i5-4210U CPU @1.70GHz 2.40GHz processors, 8.00 GB RAM and MATLAB R 2017a on WANG and SIMPSON data sets of image. Features vectors of these images are calculated using LBP and proposed m-LBP techniques. Euclidian distance and Manhattan distance are used to find the similarity difference between the query image and image dataset. We consider 5 images as the number of output images. The output image, which is at the lowest distance from the input query image, is the most similar. Precision and recall values are also calculated using eq. 9 and eq. 10. Precision is the ratio of the total number of relevant images retrieved from the dataset and the total number of images in the dataset. The recall is defined as the ratio of the total number of relevant pictures extracted from the dataset and the total number of related images present in the dataset.
𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 = 𝑇𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑟𝑒𝑙𝑒𝑣𝑎𝑛𝑡 𝑖𝑚𝑎𝑔𝑒𝑠 𝑟𝑒𝑡𝑟𝑖𝑒𝑣𝑒𝑑 𝑓𝑟𝑜𝑚 𝑡𝑒 𝑑𝑎𝑡𝑎𝑠𝑒𝑡
𝑇𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑖𝑚𝑎𝑔𝑒𝑠 𝑖𝑛 𝑡𝑒 𝑑𝑎𝑡𝑎𝑠𝑒𝑡 (9) 𝑅𝑒𝑐𝑎𝑙𝑙 = 𝑇𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑟𝑒𝑙𝑒𝑣𝑎𝑛𝑡 𝑖𝑚𝑎𝑔𝑒𝑠 𝑟𝑒𝑡𝑟𝑖𝑒𝑣𝑒𝑑 𝑓𝑟𝑜𝑚 𝑡𝑒 𝑑𝑎𝑡𝑎𝑠𝑒𝑡
𝑇𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑟𝑒𝑙𝑒𝑣𝑎𝑛𝑡 𝑖𝑚𝑎𝑔𝑒𝑠 𝑝𝑟𝑒𝑠𝑒𝑛𝑡 𝑖𝑛 𝑡𝑒 𝑑𝑎𝑡𝑎𝑠𝑒𝑡 (10)
4.1 Used Datasets
The dataset 1 (WANG Dataset) contains ten categories of 100 images, each of size 50 x 50 figure 4 (i), and the dataset 2 (SIMPSON Data set) includes ten categories of 50 images each of size 50 x 50 figure 4(ii). Both the datasets contain different categories of images. All the images are colored images. The different categories of images available in used data set 1 and data set 2 are described in table 1. In the dataset one, there are a total of ten categories, namely, African, beaches, mountains, buses, dinosaurs, elephants, rose, horse, snowy hills, and foods. In the data set 2, there are a total of ten categories, namely Abraham Grampa, Apu, Bart, Charles, Chief, Comic book guy, Edna Krabappel, Homer, Kent Brockman, and Krusty.
Figure 4(i)
(a) (b) (c) (d) (e)
(f) (g) (h) (i) (j)
Figure 4(ii)
(a) (b) (c) (d) (e)
(f) (g) (h) (i) (j)
Fig. 4: (i) and (ii) Image (a) to (j) presenting different categories of images used in dataset 1 and dataset 2(ii)
Table 1: Different types of classes and groups in used dataset 1 and data set 2
Sr. No. Class # Image Category Data Set 1 Image Category Data Set 2
1 Class 1 African Abraham Grampa
2 Class 2 Beach Apu
3 Class 3 Monument Bart
4 Class 4 Buses Charles
5 Class 5 Dinosaurs Chief
6 Class 6 Elephant Comic book guy
7 Class 7 Rose Edna Krabappel
8 Class 8 Horse Homer
9 Class 9 Snowy hills Kent Brockman
10 Class 10 Food Krusty
Fig. 5. (a) and (c) represents the precision and recall values curves and (b) and (d) represents the output images for data set 1 using Euclidean distance and Manhattan distance respectively using LBP technique. In Fig. 5(b) and fig. 5(d) the first image represents the query image and the remaining images show the retrieved images for the query image
(a) (b)
(c) (d)
Fig. 5. (a) and (c) represents the precision and recall values curves and (b) and (d) represents the output images for data set 1 using Euclidean distance and Manhattan distance respectively using LBP technique.
Fig. 6. (a) and (c) represents the precision and recall values curves and (b) and (d) represents the output images for data set 1 using Euclidean distance and Manhattan distance respectively using the m-LBP technique. In Fig. 6(b) and fig. 5(d) the first image represents the query image and the remaining images show the retrieved images for the query image
(a) (b)
(c) (d)
Fig. 6. (a) and (c) represents the precision and recall values curves and (b) and (d) represents the output images for data set 1 using Euclidean distance and Manhattan distance respectively using the m-LBP technique.
Fig. 7. (a) and (c) represent the precision and recall values curves, and (b) and (d) represents the output images for data set 2 using Euclidean distance and Manhattan distance respectively using the LBP technique. In Fig. 7(b) and fig. 5(d), the first image represents the query image, and the remaining images show the retrieved images for the query image.
(a) (b)
(c)
(d)
Fig. 7. (a) and (c) represents the precision and recall values curves, and (b) and (d) represents the output images for data set 1 using Euclidean distance and Manhattan distance respectively using the LBP technique.
Fig. 8. (a) and (c) represent the precision and recall values curves, and (b) and (d) represents the output images for data set 2 using Euclidean distance and Manhattan distance respectively using the m-LBP technique. In Fig. 8(b) and fig. 5(d), the first image represents the query image, and the remaining images show the retrieved images for the query image.
(a) (b)
(c)
(d)
Fig. 8. (a) and (c) represent the precision and recall values curves, and (b) and (d) represents the output images for data set 2 using Euclidean distance and Manhattan distance respectively using the m-LBP technique.
4.3 Time of Execution
Table 2 shows the time of execution of the LBP technique and proposed the m-LBP method on both the data sets using Euclidean distance and Manhattan distance. Results show that there is a significant improvement of approx. 42% reduction in time of execution using proposed m-LBP.
Table 2: Time of execution for two different techniques on both the data sets
Sr. No. Technique Data Set Distance Measure Time of Execution(Sec)
1 LBP
WANG
Euclidean Distance 227.6880
2 Manhattan Distance 249.2490
3 MS-LBP Euclidean Distance 129.9170
4 Manhattan Distance 132.3750
5 LBP
SIMPSON
Euclidean Distance 212.9840
6 Manhattan Distance 226.4650
7 m-LBP Euclidean Distance 110.3540
8 Manhattan Distance 122.1430
Graphical Representation of the time of execution and technique used along with datasets and distance measurement technique is shown in figure 9.
Fig 9: Time of performance of the proposed method and LBP technique for both the dataset using different distance measure techniques
5 Conclusion
In this paper, we proposed a modified-LBP (m-LBP) to extract discriminative features for the CBIR system. The proposed m-LBP overcome the shortcoming of traditional LBP by using the 4 most significant bits (MSB). Experiments have been performed on two datasets, namely WANG and SIMPSON images, using Euclidean and Manhattan distance as similarity matrices. The m-LBP has also been proved to be effective in terms of execution time. Finally, the results are compared with existing techniques, where the proposed algorithm outperforms. In the future, the method can be integrated with other CBIR descriptors to improve the performance of the system.
References
[1] Wang, X., Han, T. X., & Yan, S. (2009, September). An HOG-LBP human detector with partial occlusion handling. In 2009 IEEE 12th international conference on computer vision (pp. 32-39). IEEE.
[2] Guo, Z., Zhang, L., & Zhang, D. (2010). Rotation invariant texture classification using LBP variance (LBPV) with global matching. Pattern recognition, 43(3), 706-719.
[3] Zhang, L., Chu, R., Xiang, S., Liao, S., & Li, S. Z. (2007, August). Face detection based on multi-block LBP representation. In International conference on biometrics (pp. 11-18).
Springer, Berlin, Heidelberg.
[4] Brahnam, S., Jain, L. C., Nanni, L., & Lumini, A. (Eds.). (2014). Local binary patterns: new variants and applications (Vol. 2). Berlin: Springer.
[5] N. Arora, A. Ashok, and S. Tiwari (2019). Similar image retrieval based on texture feature vector using Local Octal and Local Hexadecimal Pattern and comparison with Local Binary Pattern. Journal of Mechanics of Continua and Mathematical Science, (vol.14, no.4, pp.558- 578)https://doi.org/10.26782/jmcms.2019.08.00046
[6] Ylioinas, J., Hadid, A., & Pietikäinen, M. (2012, November). Age classification in unconstrained conditions using LBP variants. In Proceedings of the 21st International Conference on Pattern Recognition (ICPR2012) (pp. 1257-1260). IEEE.
[7] Nanni, L., Lumini, A., & Brahnam, S. (2012). Survey on LBP based texture descriptors for image classification. Expert Systems with Applications, 39(3), 3634-3641.
0 50 100 150 200 250 300
Euclidean Distance
Manhattan Distance
Euclidean Distance
Manhattan Distance
Euclidean Distance
Manhattan Distance
Euclidean Distance
Manhattan Distance
WANG SIMPSON
LBP Proposed Technique LBP Proposed Technique
Time of Execution(Sec)
[8] Hadid, A., Ylioinas, J., Bengherabi, M., Ghahramani, M., & Taleb-Ahmed, A. (2015). Gender and texture classification: A comparative analysis using 13 variants of local binary patterns. Pattern Recognition Letters, 68, 231-238.
[9] Bedi, A. K., Sunkaria, R. K., & Randhawa, S. K. (2018, December). Local Binary Pattern Variants: A Review. In 2018 First International Conference on Secure Cyber Computing and Communication (ICSCCC) (pp. 234-237). IEEE.
[10] A. Alaknanda, A. Nitin: „Content-based image retrieval using Histogram and LBP,‟
International Journal of Communication System and Network Technology, vol. 5, No. 1, 2016, pp. 50-65
[11] Ahonen, T., Hadid, A., & Pietik$\$$\$"ainen, M. (2004). Face recognition with local binary patterns. European Conference on Computer Vision, 469–481. https://doi.org/10.1007/978-3- 540-24670-1_36
[12] Celik, T., & Tjahjadi, T. (2009). Multiscale texture classification using dual-tree complex wavelet transform. Pattern Recognition Letters, 30(3), 331–339.
https://doi.org/10.1016/j.patrec.2008.10.006
[13] Chakraborty, S., Singh, S. K., & Chakraborty, P. (2017). Local directional gradient pattern: a local descriptor for face recognition. Multimedia Tools and Applications, 76(1), 1201–1216.
[14] Dubey, S. R., Singh, S. K., & Singh, R. K. (2016). Local Bit-Plane Decoded Pattern: A Novel Feature Descriptor for Biomedical Image Retrieval. IEEE Journal of Biomedical and Health Informatics, 20(4), 1139–1147. https://doi.org/10.1109/JBHI.2015.2437396
[15] Arora N., et al.: "Efficient Image Retrieval through Hybrid Feature Set and Neural network,"
International Journal of Image, Graphics and Signal Processing(IJIGSP), Vol.11, No.1, pp.
44-53, 2019.
[16] Hamouchene, I., & Aouat, S. (2014). A New Texture Analysis Approach for Iris Recognition.
AASRI Procedia, 9, 2–7. https://doi.org/10.1016/j.aasri.2014.09.002
[17] He, Y., Sang, N., & Gao, C. (2012). Multi-structure local binary patterns for texture classification. Pattern Analysis and Applications, 1–13. https://doi.org/10.1007/s10044-011- 0264-4
[18] Kokare, M., Biswas, P. K., & Chatterji, B. N. (2006). Rotation-invariant texture image retrieval using rotated complex wavelet filters. IEEE Transactions on Systems, Man, and
Cybernetics, Part B: Cybernetics, 36(6), 1273–1282.
https://doi.org/10.1109/TSMCB.2006.874692
[19] Kokare, M., Biswas, P. K., & Chatterji, B. N. (2007). Texture image retrieval using rotated wavelet filters. Pattern Recognition Letters, 28(10), 1240–1249.
https://doi.org/10.1016/j.patrec.2007.02.006
[20] Nanni, L., Lumini, A., & Brahnam, S. (2010). Local binary pattern variants as texture descriptors for medical image analysis. Artificial intelligence in medicine, 49(2), 117-125.
[21] Kazak, N., & Koc, M. (2018). Some variants of spiral LBP in texture recognition. IET Image Processing, 12(8), 1388-1393.
[22] Guo, Z., Zhang, L., & Zhang, D. (2010b). Rotation invariant texture classification using LBP variance (LBPV) with global matching. Pattern Recognition, 43(3), 706–719.
https://doi.org/10.1016/j.patcog.2009.08.017
[23] N. Arora, A. Ashok, S. Tiwari, “Modified Local Binary Pattern Scheme using Row, Column and Diagonally aligned Pixel‟s Intensity Pattern” International Journal of Innovative Technology and Exploring Engineering (IJITEE), vol. 8, no. 5, pp. 771-779, March 2019
Authors
Mamta Martolia is pursuing a Ph.D. at Uttarakhand Technical University, Dehradun, India. Her current research interests include Cryptography, Network Security, Software Engineering.
Nilesh Dhanore, is working as an Assistant Professor in the department of Electronics and Communication Engineering at Jhulelal Institute of Technology, Nagpur, India. His research interests are in Electronics & Communication, VLSI and Artificial Intelligence.
Anupam Singh, is Assistant Professor-senior Scale in Department of Informatics, School of Computer Science, University of Petroleum, Dehradun. He is pursuing a Ph.D.
from Dr. APJ Abdul Kalam Technical University Lucknow (Formerly UPTU Lucknow). He has done B. Tech. in 2004, M. Tech. in 2011 from Uttar Pradesh Technical University Lucknow. His areas of interest encompass Formal Methods, Distributed System, and Database System.
Vivek Shahare is an Assistant Professor in the School of Computer Science, University of Petroleum & Energy Studies, Dehradun. He has received his B.Tech from Government College of Engineering, Amravati in 2012, and M.Tech from Visvesvaraya National Institute of Technology (VNIT), Nagpur, in 2016. His research interest includes Bioinformatics and Cryptography.
Nitin Arora is pursuing a Ph.D. in Engineering at the Indian Institute of Technology Roorkee, India. His research interest includes Image Processing and Machine learning.
