International Journal of Electrical and Computer Engineering (IJECE) Vol.
No.
June 2016, pp.
ISSN: 2088-8708.
DOI: 10.
11591/ijece.
Stone Image Classification Based on Overlapped 5-bit T-Patterns Occurrence on 5-by-5 Sub Images Palnati Vijay Kumar.
Pullela S V V S R Kumar.
Nakkella Madhuri.
M Uma Devi Department of Computer Science & Engineering.
Aditya College of Engineering.
Surampalem.
Andhra Pradesh.
India
Article Info
ABSTRACT
Article history:
Texture classification is widely used in understanding the visual patterns and has wide range of applications.
The present paper derived a novel approach to classify the stone textures based on the patterns occurrence on each sub The present approach identifies overlapped nine 5 bit T-patterns (O5TP) on each 5y5 sub window stone image.
Based the number of occurrence of T-patterns count the present paper classify the stone images into any of the four classes i.
brick, granite, marble and mosaic stone The novelty of the present approach is that no standard classification algorithm is used for the classification of stone images.
The proposed method is experimented on Mayang texture images.
Brodatz textures.
Paul Bourke color images.
VisTex database.
Google color stone texture images and also original photo images taken by digital camera.
The outcome of the results indicates that the proposed approach percentage of grouping performance is higher to that of many existing approaches.
Received Sep 20, 2015 Revised Nov 6, 2015 Accepted Nov 26, 2015 Keyword:
Grey level image Image classification Stone images Texture Classification T-patterns Copyright A 2016 Institute of Advanced Engineering and Science.
All rights reserved.
Corresponding Author:
Dr.
Pullela S V V S R Kumar.
Departement of Computer Science & Engineering.
Aditya College of Engineering.
Surampalem.
Andhra Pradesh.
India.
Email: pullelark@yahoo.
INTRODUCTION
TEXTURE analysis and categorization are important for the interpretation and understanding of real-world visual patterns.
Texture classification has a wide variety of prospective applications .
such as regions classification in satellite images .
, defects detection in industrial surface inspection .
, and classification of pulmonary disease .
, diagnosis of leukemic cells in medical image .
and breast cancer classification .
Texture analysis and classification is majorly achieved in one of the two ways, i.
statistical approach and structural method.
Statistical approach mainly concentrates on the stochastic things of the spatial distribution of gray levels in an image.
Generally for finding the characteristics, co-occurrence matrix is frequent.
From the co-occurrence matrix set of textural features extracted and these features are widely used to extract textural information from digital images .
In structural approach, texture is considered as a repetition of some primitives.
For texture classification and characterization, these methods have been applied by several authors and achieved success to a certain degree .
Characterization and classification of textures is an important step in the study of patterns on texture The textures are characterized and classified recently by various pattern approach methods: edge direction movements .
, long linear patterns .
and preprocessed images .
Marble texture description .
, avoiding Complex Patterns .
Texture images are also described and classified by using various wavelet transforms techniques: one based on statistical parameters .
and another one based on primitive patterns .
Sasi Kiran et.
has proposed a method called Wavelet based Histogram on texton patterns (WHPT) and grouped the stone textures into four categories.
The WHTP method got average % of grouping Journal homepage: http://iaesjournal.
com/online/index.
php/IJECE IJECE
A 1153
ISSN: 2088-8708
Dr.
U Ravi Babu et.
has proposed a method for stone textures classification into four groups.
In this method also used patterns approach on grey-to-grey level preprocessed images.
This method also 15% as group classification, but this method is applied only for grouping stone textures into four Suma Latha et.
has proposed a method called LBP-High-Symmetry (LBP-HS) for recognition of stone textures.
This approach is also patterned approach for stone texture recognition.
The LBP-HS method got 92% of recognition only.
Sujatha B et al .
proposed a method called Texton and Texture Orientation Co-occurrence Matrix (T&TO-CM) for the classification of textures.
The proposed method achieved only 93% of classification rate.
In most approaches, which have been offered so far, researchers have tried to analyze and describe texture based on overlapped alphabet patterns for stone image classification.
The proposed method put forward the pattern approach for grouping the stone textures into four classes.
The high accuracy in texture classification in the results shows the quality of offered approach.
The present paper proposes an approach for stone textures classification based on occurrences of overlapped T-patterns on each 5y5 sub-images.
The reminder of this paper is organized as follows: Section two describes to the identification of Overlapped 5-bit T-Patterns (O5TP) on the grey level image.
Section three is related to deriving an algorithm for grouping the stone texture and analyses the results and finally, the conclusion included.
PROPOSED METHOD
Identification of Overlapped 5-bit T-Patterns on Each 5y5 sub-stone image The proposed method O5TP consists of 4 steps.
In step 1, convert the each stone texture color into the grey level image by using 7-bit binary code quantization method.
Identify the 5-bit T-patterns in each 5y5 window of the stone texture image in step 2.
In step 3, count the occurrences of T-patterns.
Finally, based on the number of T-patterns derive a new algorithm for classification.
The block diagram of the entire procedure is shown in Figure 1.
Stone Color Quantization Count the T-patterns Grey Identify Overlapped 5bit T-patterns User defined Classification Figure 1.
Block diagram of the proposed stone image classification Step 1: Color to Grey Scale conversion The Color image is nothing but a color channels.
Most digital images are comprised of three separate color channels: a red channel, a green channel, and a blue channel.
Grey scale means many shades .
from black to white.
Generally, 7 ways are available to convert the color image into gray scale image averaging method, luma method, and De-saturation method.
Custom #of gray shades method, horizontal error-diffusion dithering method.
Single color channel and Single color channel method.
In this paper utilized Custom # of gray shades method.
Custom # of gray shades method: this allows the user to specify how many shades of gray the resulting image will use.
This value can be between 2 and 256 is accepted.
If it is 2, the resultant image contains 2 shades i.
black-and-white image, while 256 gives an image consists of 256 shades.
The proposed method uses 8-bit color channels.
So, maximum shades are only 256.
In this paper uses 128 shades.
Any grayscale conversion algorithm is a three-step process:
Catch the green, red and blue values of a pixel Convert those three values into a single gray value Replace the three values with the new gray value Elaborated algorithm for Grey Scale conversion Step 1: Exchange threshold value = 255/(Number Of Shades-.
Step 2: mean value = (Red Green Blu.
/ 3 Step 3: Gray = Integer .
ean value / exchange threshold valu.
* exchange threshold value Step 2: identification of 5-bit T-patterns each 5y5 grey-level stone sub image The 5y5 sub image values are represented as P1.
P2.
A P9.
P10.
P11.
A P24.
P25.
The pixel position of the each 5y5 sub window was shown in Figure 2.
Stone Image Classification Based on Overlapped 5 bit T Patterns Occurrence on A.
(Palnati Vijay Kuma.
A ISSN: 2088-8708 Figure 2.
Pixel positions in 5y5 grey level facial sub image In the proposed method consider the all possible T-pattern formed using 5-bits.
The first T-pattern formed using 5 pixel P1.
P2.
P3.
P7, and P12 and the second T-pattern formed using 5 pixel P2.
P3.
P4.
P8, and P13 and so on.
From first row 3 T-patterns are formed.
From the Second row, another 3 T-patterns are From the 3rd row another set of 3 patterns formed.
Totally, nine overlapped 5-bit T-patterns are possible in each 5y5 sub window.
Figure 3 shows the possible overlapped T-patterns on each 5y5 window.
Figure 3.
Overlapped 5-bit T-patterns on each 5y5 window Step 3: Count the T-patterns Count the frequency occurrence of the considered patterns in each 5y5 sub window on the stone texture adds these values to the feature vector.
Step 4: Classification of stone texture images Based on frequency occurrences of overlapped 5-bit T-pattern in each 5y5 sub window on the stone texture image is classified as one of the four categories i.
Marble.
Mosaic and Granite.
RESULTS AND DISCUSSION
To find the effectiveness of the proposed approach (O5TP) carried out the experiments on mixed stone textures Dataset which consists of various brick, granite, marble, and mosaic stone textures collected from Mayang.
Google.
CuRet VisTex, and Paul Bourke data-base and also from floor images taken by camera with the resolution of 256y256.
The images used in this experiment is 1880 i.
480 images from Mayang database, 410 images from Paul Bourke database, 160 images from VisTex database, 130 images from CuRet database 300 images from Google database and 400 images from scanned photo graphs.
Figure 4 shows the some of the stone textures used in this paper to evaluate the efficiency of the proposed approach.
Some of the frequency of occurrence of overlapped 5-bit T-pattern (O5TP) of marble, mosaic, granite, and brick texture dataset images are listed-out in table 1 to 4 respectively.
From Tables 1 to 5 designs a classification graph of five categories of stone images is shown in Figure 5.
IJECE Vol.
No.
June 2016 : 1152 Ae 1160
IJECE
A 1155
ISSN: 2088-8708
Figure 4.
some of the stone images used in this method with resolution of 256y256 Table 1.
Overlapped 5-bit T-patterns occurrences of Brick textures Image Name Brick.
Brick.
Brick.
Brick.
Brick.
Brick.
Brick.
Brick.
Brick.
Brick.
Brick.
Brick.
Brick.
Brick.
Brick.
Brick.
Brick.
Brick.
Brick.
Brick.
Pattern1 Image Name blue_granite blue_pearl blue_topaz brick_erosion canyon_black dapple_green ebony_oxide giallo_granite gosford_stone interlude_haze mesa_twilight mesa_verte pietro_nero russet_granite Pattern1 Pattern2 Pattern3 Pattern4 Pattern5 Pattern6 Pattern7 Pattern8 Pattern9 Total Table 2.
Overlapped 5-bit T-patterns occurrences of Granite textures Pattern2 Pattern3 Pattern4 Pattern5 Pattern6 Pattern7 Pattern8 Pattern9 Total Stone Image Classification Based on Overlapped 5 bit T Patterns Occurrence on A.
(Palnati Vijay Kuma.
A ISSN: 2088-8708 Table 3.
Overlapped 5-bit T-patterns occurrences in Horizontal Direction of Marble textures Image Name canyon_blue curry_stratos flinders_blue flinders_green forest_boa forest_stone green_granite grey_stone Pattern1 Pattern2 Pattern3 Pattern4 Pattern5 Pattern6 Pattern7 Pattern8 Pattern9 Total Table 4.
Overlapped 5-bit T-patterns occurrences in Horizental Direction of Mosaic textures Image Name concrete_bricks_170756 concrete_bricks_170757 concrete_bricks_170776 crazy_paving_5091370 crazy_paving_5091376 crazy_tiles_130356 crazy_tiles_5091369 dirty_floor_tiles_footprints_2564 dirty_tiles_200137 floor_tiles_030849 grubby_tiles_2565 kitchen_tiles_4270064 moroccan_tiles_030826 moroccan_tiles_030857 mosaic_tiles_8071010 mosaic_tiles_leaf_pattern_201005060 mosaic_tiles_roman_pattern_201005034 motif_tiles_6110065 ornate_tiles_030845 repeating_tiles_130359 Pattern Pattern Pattern Pattern Pattern Pattern Pattern Figure 5.
The proposed method generated classification graph IJECE Vol.
No.
June 2016 : 1152 Ae 1160
Pattern
Pattern
Total
IJECE
ISSN: 2088-8708
A 1157
Figure 6.
Generated Classification Graph of marble and Granite stone textures The generated graph shown in figure 5 doesnAot clearly indicate the granite and marble because of the occurrences of 5-bit T-patterns are less compare to other two groups.
So, separate graph is generated for the occurrences of 5-bit T-patterns in marble and granite stone image.
The generated classification graph for marble and granite is shown in figure 6.
From the tables 1 to 4 and the classification graphs of Figure 5and 6 assign an exact and specific classification of color stone images using rate of recurrences of overlapped 5-bit T-patterns.
A new algorithm is derived for classification among these four classes i.
Granite.
Marble.
Mosaic, and Brick group of stone textures based on the above table values and generated graph.
The rate of occurrences of 5-bit T-patterns is dependent on the dimension of the texture that means when dimensions of the image changed.
the rate of occurrences is also changed.
To avoid such problems the present paper derived a classification algorithm independent of the image size.
This algorithm categorizes the stone textures in to four groups irrespective of their dimensions.
The derived algorithm uses 256y256 dimension as a bench
If the rate of occurrences of the test image cataract within the range of minimum to maximum quantity of occurrences of two and four transitions of a fastidious stone then test image is categorized as a particular
Algorithm 1: Stone texture classification based on Overlapped 5-bit T-Patterns Let Sum of occurrences of Overlapped 5-bit T-Patterns SOTP
START
if SOTP<= Test image texture group is categorized as GRANITE class
Otherwise if SOTP <= Test image texture group is categorized as MARBLE class
Otherwise if SOTP <= Test image texture group is categorized as MOSAIC class
Otherwise if SOTP<= ( Test image texture group is categorized as BRICK class
Otherwise Test image texture group is categorized as UNKNOWN class
STOP
COMPARISON BETWEEN PROPOSED METHOD AND OTHER EXISTING METHODS
The proposed method is compared with Wavelet based Histogram on Texton Patterns (WHTP) .
, which is used to classify the stone texture images into four categories by using wavelet based texton pattern histogram and texton feature evolution method .
, which is used to classify the images into four groups Stone Image Classification Based on Overlapped 5 bit T Patterns Occurrence on A.
(Palnati Vijay Kuma.
A ISSN: 2088-8708 based on rate of occurrences of texton patterns.
The proposed method is also compared with other existing method like Syntactic Pattern on 3D method .
in which stone textures are classified into four categories based on the occurrence of systematic patterns.
It is clearly obvious that, the proposed method show signs of a high classification rate than the existing methods.
The percentage mean classification rate for the proposed method and other existing methods are represented in Table 5.
The graphical representation of the percentage mean classification rate for the proposed method and other existing methods are shown in Figure 7.
The Table 5 and Figure 7 shows the mean percentage classification of original images Google and scanned image.
The mean percentage classification of proposed method and other existing methods of various databases are represented in Table 6 and graphical representation is shown in Figure 8.
Table 5.
Mean percentage classification results of the proposed method and other existing methods Image Dataset Original Google Scanned Average Syntactic Pattern on 3D Wavelet based Histogram on Texton Patterns Texton Feature Detection Proposed Method Figure 7.
Classification chart of proposed method with other existing methods Table 6.
Mean percentage classification rates of the proposed method and other existing methods Image Dataset VisTex Texture Images Taken by Camera CuReT Mayang Paul Bourke Syntactic Pattern on 3D method Wavelet based Histogram on Texton Patterns Texton Feature Detection Proposed Method Figure 8.
Mean percentage classification chart of the proposed method and other existing methods IJECE Vol.
No.
June 2016 : 1152 Ae 1160
IJECE
ISSN: 2088-8708
A 1159
No standard classification algorithm is used to test the data base.
The novelty of the proposed method is that the proposed technique is applied on huge dataset.
Even though it is applied on huge dataset it gives good results when compare with the other existing methods.
Still, no such technique is available to apply on large dataset.
CONCLUSION
The present paper derived a new approach called Overlapped 5-bit T-Patterns (O5TP) for stone texture classification.
The present paper considered Nine 5-bit T-patterns on each 5y5 sub image without losing the information about the image for texture analysis of the grey level image.
The novelty of the proposed method is no standard classification algorithm is used for classification of stone textures.
Proposed method is tested by using large set data base and got high % of group classification i.
the strength of the proposed method.
When compare with the other existing method gives more accurate and precise classification results.
The O5TP is computationally inexpensive.
The experimental results clearly indicate the efficacy of the proposed O5TP over the various existing methods.
REFERENCES