# Introduction nalysis of textures is a fundamental research topic in the area of computer vision and has many potential applications, for example, in industrial surface inspection, remote sensing, and biomedical image analysis. Classification refers to as assigning a physical object or incident into one of a set of predefined categories. Many texture classification problems usually require the computation of a large amount of texture features in order to characterize their associated patterns. This implies that texture classifiers frequently combine big sets of features without taking into account their relevance and redundancy. Thus, lowering the dimensionality of a feature set is necessary for preserving the most relevant features and it reduces the computational cost derived from unnecessary features [1,2,3,34,35]. Numerous algorithms of textural features extraction have been presented during the past decades [4,5]. Textures are classified recently by various methods: preprocessed images [34], long linear patterns [35], edge direction movements [21], avoiding complex patterns [10], marble texture description [36], skeleton extraction of texture [7], long linear patterns using wavelets [8] wavelet transform [8,9,10]. and Gabor filters [11]. More recently, the local-binary-pattern (LBP) operator [12,13,14] is used for texture classification. LBP operator is a statistical texture descriptor of the characteristics of the local structure. LBP provides a unified description including both statistical and structural characteristics of a texture patch, so that it is more powerful for texture analysis. The concept of LBP is also extend in applications such as face recognition and age classification [15,16,17], industrial visual inspection [18,19], segmentation of remote-sensing images [20], and classification of real outdoor images [21]. An efficient nonparametric methodology for texture analysis based on magnitude LBP (MLBP) [22,23,24,25,26] is recently proposed and it has been made into a powerful measure of image texture, in terms of accuracy and computational complexity in many empirical studies. To address the connectivity limitations of LBP and MLBP, we propose a matrix called Triangular Neighborhood Matrix (TNM), which generates 2×2 texton patterns. A fuzzy member ship is introduced on TNM to extract local texture information efficiently. The present paper derived run length matrix on the proposed scheme and evaluated runlength features for efficient, precise and accurate classification of textures. A ( D D D D ) F 2012 # Year The rest of the paper is organised as follows. Section 2 describes the proposed method. Section 3 describes the results and discussions and conclusions are given in section 4. pattern (FTGP), triangular neighbor pixels local greylevel matrix (LGM). # II. # Methodology # Derivation of TNM (Triangular Neighborhood Matrix) The present paper derived FTGP to overcome the disadvantages of LBP and other local binary approaches. Runlength features are evaluated on FGTP for a precise classification in 5 steps. Step 1: Formation of Local Grey level Matrix (LGM): A neighborhood of 3×3 pixels is denoted by a set containing nine elements: P= {P1, P1 ...P9}, here P5 represents the intensity value of the central pixel and remaining value are the intensity of neighboring pixels as shown in Fig. 1(a). The Local Grey level Matrix (LGM) values of the neighboring pixels (LGMPi) are obtained by evaluating the absolute difference between the neighboring pixel and the gray value of the central pixel, as described by the Equation (1) as shown in Fig. 1. P 1 P 2 P 3 LGMP 1 LGMP 2 LGMP 3 P 4 P 5 P 6 LGMP 4 LGMP 5 LGMP 6 P 7 P 8 P 9 LGMP 7 LGMP 8 LGMP 9 Where LGMP i is the obtained grey value of the pixel P i of the LGM. The equation 1 demonstrates that always LGMP 5 value (central pixel value) will be always zero. Step 2: Generation of Triangular Neighborhood Matrix (TNM) from LGM of step 1: The 2 x 2 TNM is generated from LGM by taking the average value of the Triangular Neighbor Pixels (TNP) of the 3 x 3 LGM as shown in figure 3 and as given in equation 2,3, 4 and 5 . The triangular neighbors are considered because the central pixel of LGM is always zero. That is one need not necessary to consider this. ?????? 1 = (LGMP 1 + LGMP 1 + LGMP 1 )3?????? 2 = ?LGMP 2 + LGMP 3 + LGMP 6 ?3?????? 3 = (LGMP 4 + LGMP 7 + LGMP 8 )3?????? 4 = (LGMP 6 + LGMP 8 + LGMP 9 )3 LGMP 1 LGMP 2 LGMP 3 LGMP 4 LGMP 5 LGMP 6 TNP 1 TNP 2 LGMP 7 LGMP 8 LGMP 9 Fuzzy logic has certain major advantages over traditional Boolean logic when it comes to real world applications such as texture representation of real images. LBP patterns are formed and counted from 0's and 1's. However, the dangerous situation of LBP is that even if the difference is minimum let us say 1 or maximum i.e. 255, it converts it into 1. That is LBP treats even the difference of 1 and 255 as homogeneous. This clearly indicates the patterns of LBP will never gives totally useful and significant information. The above property misuses the power of LBP method. To address this in the proposed method fuzzy member ship is introduced. The aim of fuzzy approach in forming FTGP is to extract local texture information from TNM pixels for representing the texture information accurately. To deal accurately with the regions of natural images even in the presence of noise and the different processes of caption and digitization FTGP is introduced on TNM. For example, even if the human eye perceives two neighboring pixels as equal, they rarely have exactly the same intensity values. The fuzzy patterns are chosen in the present paper because, recently, fuzzy based methods have been used in texture analysis and in image segmentation [28,29]. The FTGP consists of fuzzy patterns with 5 values {0, 1, 2, 3 and 4} instead of two patterns of LBP. Though the present paper considers five possible fuzzy grey level values, but at any time only a maximum of four fuzzy patterns will appear because the FTGP is a 2 x 2 matrix. In LBP binary patterns are evaluated by comparing the neighboring pixels with central pixel. The FTGP are derived by comparing the each pixel of the 2 x 2 TNM with the average pixel values of the TNM. The FTGP representation is shown in Fig. 4. The following Eqn. ( 6) is used to determine the elements, FTGP i of the TNM. For example, the process of evaluating FTGP from a sub TNM image of 2 x 2 is shown in Fig. 5. In this example x and y are chosen as v 0 /2 and 3v o /2 respectively. The membership values of FTGP neighboring pixels are useful for characterization of textures. To address this difficulty the present approach derived Run length matrix (RLM) on the FTGP of the image. # Definition of the Run-Length Matrices: Galloway proposed the use of a run-length matrix for texture feature extraction [12]. For a given texture image, a runlength matrix P(i; j) is defined as the number of runs with fuzzy value i and run length j. Various texture features can then be derived from this run-length matrix. For a given image, the proposed method defines a RLM (i,j) on FTGP as number of runs starting from location (i,j) of the FTGP image. The proposed method derived five different RLM-FTGP. The RLM-FTGP 0 , RLM-FTGP 1 , RLM-FTGP 2 RLM-FTGP 3 and RLM-FTGP 4 contain the run length values for zero, one, two, three and four. Step 5: Extraction of Texture Features on RLM -FTGP: Many researchers used three sets of texture features from RLM for texture classification. The first set of RLM Features (RF) is Traditional Run-Length Features. The five original features of run-length statistics derived by Galloway [27] are Short Run Emphasis (SRE), Long Run Emphasis (LRE), Gray-Level Non uniformity (GLN), Run Length Non uniformity (RLN), and Run Percentage (RP) are described by the Equation (8) to Equation (12). Chu et al. [30] proposed another set of two new features, such as Low Gray-Level Run Emphasis (LGRE), and High Gray-Level Run Emphasis (HGRE) are described in Equation ( 13) to Equation (14). In a recent study, Dasarathy and Holder [31] described another set of four feature extraction functions following the idea of joint statistical measure of gray level and run length, as follows: Short Run Low Gray-Level Emphasis (SRLGE), Short Run High Gray-Level Emphasis (SRHGE), Long Run Low Gray-Level Emphasis (LRLGE), and Long Run High Gray-Level Emphasis (LRHGE) are described in Equation (15) to Equation (18). The novelty of the present study is it evaluated the first five RFs as described in equations from 8 to 12 for efficient classification purpose on FTGP. For a comparative analysis the present paper also evaluated all the features for classification purpose. ?????? = 1 ?? ?? ? ? ??(??,?? ) ?? 2 ?? ?? =1 ?? ??=1 (8) ?????? = 1 ?? ?? ? ? ??(??, ??) * ?? 2 ?? ?? =1 ?? ??=1 ?????? = 1 ?? ?? ? ? ??(??,?? ) ?? 2 ?? ?? =1 ?? ??=1 (10) ?????? = 1 ?? ?? ? ?? ??(??, ??) ?? ?? =1 ? 2 ?? ??=1 (11) ???? = n r n p(12) In the above equations, n r is the total number of runs and n p is the number of pixels in the image. ???????? = 1 ?? ?? ? ? ??(??,?? ) ?? 2 ?? ?? =1 ?? ??=1 (13) ???????? = 1 ?? ?? ? ? ??(??, ??) * ?? 2 ?? ?? =1 ?? ??=1 (14) ?????????? = 1 ?? ?? ? ? ??(??,?? ) ?? 2 * ?? 2 ?? ?? =1 ?? ??=1 (15) ?????????? = 1 ?? ?? ? ? ??(??,?? ) * ?? 2 ?? 2 ?? ?? =1 ?? ??=1 (16) ?????????? = 1 ?? ?? ? ? ??(??,?? ) * ?? 2 ?? 2 ?? ?? =1 ?? ??=1 (17) ?????????? = 1 ?? ?? ? ? ??(??, ??) * ?? 2 ?? ?? =1 ?? ??=1 * ?? 2(18) III. # RESULTS and DISCUSSIONS Experiments are carried out to demonstrate the effectiveness of the proposed FTGP -with RF for stone I, Table II. Table 3 shows the classification rate for various group of textures by the proposed FTGP-RF with other existing methods like compound local binary pattern (CLBP) of Faisal Ahmed et.al [32] and run-length features for image classification by Yung-Kuan Chan et.al [33]. From Table 3, it is clearly evident that, the proposed FTGP-RF exhibits a high classification rate than the existing methods. The graphical representation of the percentage mean classification rate for the proposed RLM-FTGP and other existing methods are shown in Fig. 8. # Conclusion The proposed FTGP scheme reduces the overall dimension of the image while preserving the significant attributes, primitives, and properties of the local texture. The proposed RLM-FTGP overcomes the disadvantages of the previous Run length matrices for texture classification. LGM is an efficient tool that overcomes the traditional neighborhood problems. By directly using the entire run-length matrix for feature extraction, much of the texture information is preserved. The novelty of the proposed scheme is, it is proved that one need not necessary to evaluate all the RF on the FTGP for classification purpose. For a precise, significant and accurate classification, the present paper evaluated only 5 RLMF on FTGP, which reduced overall complexity. Comparisons of this new approach with the compound local binary pattern (CLBP) by Faisal Ahmed et.al [32] and run-length features for image classification by Yung-Kuan Chan et.al [33] demonstrated the supremacy of the proposed FTGP method. 1![Fig. 1 : (a) A neighborhood of 3×3 (b) obtained LGM](image-2.png "Fig. 1 :") 3![Figure 3 : Generation process of a 2 × 2 TNM from LGM (a) LGM matrix (b) TNM Step 3: Conversion of TNM in to FTGP (Fuzzy Triangular Grey level Pattern):Fuzzy logic has certain major advantages over traditional Boolean logic when it comes to real world applications such as texture representation of real images. LBP patterns are formed and counted from 0's and 1's. However, the dangerous situation of LBP is that even if the difference is minimum let us say 1 or maximum i.e. 255, it converts it into 1. That is LBP treats even the difference of 1 and 255 as homogeneous. This clearly indicates the patterns of LBP will never gives totally useful and significant information. The above property misuses the power of LBP method. To address this in the proposed method fuzzy member ship is introduced. The aim of fuzzy approach in forming FTGP is to extract local texture information from TNM pixels for representing the texture information accurately. To deal accurately with the regions of natural images even in the presence of noise and the different processes of caption and digitization FTGP is introduced on TNM. For example, even if the human eye perceives two neighboring pixels as equal, they rarely have exactly the same intensity values. The fuzzy patterns are chosen in the present paper because, recently, fuzzy based methods have been used in texture analysis and in image segmentation[28,29]. The FTGP consists of fuzzy patterns with 5 values {0, 1, 2, 3 and 4} instead of two patterns of LBP. Though the present paper considers five possible fuzzy grey level values, but at any time only a maximum of four fuzzy patterns will appear because the FTGP is a 2 x 2 matrix. In LBP binary patterns are evaluated by comparing the neighboring pixels with central pixel. The FTGP are derived by comparing the each pixel of the 2 x 2 TNM with the average pixel values of the TNM. The FTGP representation is shown in Fig.4. The following Eqn. (6) is used to determine the elements, FTGP i of the TNM.](image-3.png "Figure 3 :") 4![Fig. 4 : Fuzzy triangular grey level texture number representation](image-4.png "Fig. 4 :") 5![Fig. 5 : The process of evaluating FTGP from TNM (a) TNM (b) FTGP Step 4: Generation of Run Length Matrices on Fuzzy Texture Grey level Pattern (RLM-FTGP)The membership values of FTGP neighboring pixels are useful for characterization of textures. To address this difficulty the present approach derived Run length matrix (RLM) on the FTGP of the image.](image-5.png "Fig. 5 :") ![Classification Based on Fuzzy Triangular Greylevel Pattern and Run-Length Features classification. The present paper carried out the experiments on two Datasets. The Dataset-1 consists of various brick, granite, and marble and mosaic stone textures with resolution of 256×256 collected from Brodatz textures, Vistex, Mayang database and also from natural resources from digital camera. Some of them in Dataset-1 are shown in the Fig. 6. The Dataset-2 consists of various brick, granite, and marble and mosaic stone textures with resolution of 256×256 collected from Outtex, Paulbourke color textures database, and also from natural resources from digital camera. Some of them in Dataset-2 are shown in the Fig. 7. Dataset-1 and Dataset-2 contains 80 and 96 original color texture images respectively. For classification the proposed method initially divide the texture images into non-overlapping windows of size 32×32 and the resulting windows are then divided into two disjoint sets, one for training and one for testing. The distance classifier Euclidean distance (d) is used for classification in the present paper. The classifier computes the distance between the features for each sample and that of the texture classes and assigns the unknown sample to the texture class with the shortest distance. The classification results for each of the two Data sets are shown in Table](image-6.png "") 617![Fig. 6 : Input texture group of 9 samples of Granite, Brick, Mosaic, and Marble in Dataset-1](image-7.png "Fig. 6 : 1 Fig. 7 :") 8![Fig. 8 : Classification chart of proposed FTGP-RF with other existing methods IV.](image-8.png "Fig. 8 :") Ia2012Year20D D D D ) F(SnoTex ture NameClassificaTex tureClassification RateNametion Rate1concrete_bricks_17075694.22Brick.000195.062concrete_bricks_17075794.58Brick.000291.493concrete_bricks_17077689.64Brick.000397.284crazy_paving_509137095.2Brick.000495.95crazy_paving_509137696.56Brick.000593.396crazy_tiles_13035693.54Brick.000696.657crazy_tiles_509136995.8894.518dirty_floor_tiles_footprints_256493.17Brick.000893.259dirty_tiles_20013793.99Brick.000993.3710 floor_tiles_03084996.55Brick.001095.9611 grubby_tiles_256594.68Brick.001192.4612 kitchen_tiles_427006495.48Brick.001294.5213 moroccan_tiles_03082696.35Brick.001393.6214 moroccan_tiles_03085795.77Brick.001491.4815 mosaic_tiles_807101096.16Brick.001593.6116 mosaic_tiles_leaf_pattern_20100506094.97Brick.001692.0117 mosaic_tiles_roman_pattern_20100503490.91Brick.001794.5818 motif_tiles_611006595.34Brick.001892.4719 ornate_tiles_03084596.44Brick.001996.1320 repeating_tiles_13035990.84Brick.002095.37Results of texture classification by proposed RF on FTGP of mosaic and brick textures in Dataset-1 © 2012 Global Journals Inc. (US) Global Journal of Computer Science and Technology Volume XII Issue XV Version I Ib 2a 2b 3with other existing methodsImage DatasetCompound Local Binary Pattern (CLBP)Run-length FeaturesProposed Method (FTGP-RF)Brodatz90.2993.7996.31VisTex91.5393.5695.85Mayang92.3494.4397.32Outtex,91.5993.6396.96CUReT91.7693.4697.54Paulbourke90.9894.5696.77Average91.4193.9196.79 © 2012 Global Journals Inc. (US) ## Acknowledgment The authors would like to express their gratitude to Sri K.V.V. Satya Narayana Raju, Chairman, and K. Sashi Kiran Varma, Managing Director, Chaitanya group of Institutions for providing necessary infrastructure. Authors would like to thank anonymous reviewers for their valuable comments and Dr. G.V.S. Ananta Lakshmi for her invaluable suggestions which led to improvise the presentation quality of this paper. * Feature Selection for Classification. Intelligent Data Analysis MDash HLiu 1997 Elsevier * Toward optimal feature selection DKoller MSahami Proceedings of the 13th International Conference on Machine Learning the 13th International Conference on Machine LearningBari, Italy 1996 * Learning optimal filter representation for texture classification PZhang JPeng BBuckles International Conference on Pattern Recognition Hong Kong 2006 2 * A review of recent texture segmentation and feature extraction techniques TRReed JM HBuf CVGIP: Image Understanding 57 1993 * The Handbook of Pattern Recognition and Computer Vision MTuceryan AKJain Texture analysis CHChen LFPau PS PWang Singapore World Scientific 1998 second ed. * Texture as the basis for individual tree identification ASamal JRBrandle DSZhang Information Sciences 176 2006 * Texture Classification Based On Extraction Of Skeleton Primitives Using Wavelets U S NRaju EswarReddy VijayaKumar BSujatha Information Technology Journal 7 2008 * Employing Long Linear Patterns for Texture Classification relying on Wavelets VijayaKumar .V US N Raju ChandraSekaran V VKrishna ICGST-GVIP Journal 1687-398X 8 January 2009 * Radon transform orientation estimation for rotation invariant texture analysis KJafari-Khouzani HSoltanian-Zadeh IEEE Transactions on Pattern Analysis and Machine Intelligence 27 6 2005 * W.-LLee -CYung Chen -CYing K.-SChen * Unsupervised segmentation of ultrasonic liver images by multiresolution fractal feature vector Hsieh Information Sciences 175 2005 * A comparative study of texture measures with classification based on feature distribution TOjala MPietikäinen DHarwood Pattern Recognition 29 1996 * Multiresolution gray-scale and rotation invariant texture classification with local binary patterns TOjala MPietikäinen TMäenpää IEEE Transactions on Pattern Analysis and Machine Intelligence 24 7 2002 * Rotation-invariant texture classification using feature distribution MPietikäinen TOjala ZXu Pattern Recognition 33 2000 * Face description with local binary patterns: application to face recognition TAhonen AHadid MPietikäinen IEEE Transactions on Pattern Analysis and Machine Intelligence 28 12 2006 * Classification of child and adult based on geometric features of face using linear wavelets ChandraMohan MVijaya Kumar VSujatha B IJSIP 1 3 2010 * Adulthood classification based on geometrical facial features ChandraMohan VVijayakumar ADamodaram 2009 ICGST * Novel method of adult age classification using linear wavelet transforms ChandraMohan VVijayakumar VVenkata Krishna IJCSNS 2010 * Textured-based characterization of defects in automobile engine valves AMarzabal CTorrens AGrau Proceedings of the Ninth Symposium on Pattern Recognition and Image Processing the Ninth Symposium on Pattern Recognition and Image Processing 2001 * Supervised segmentation of textures in backscatter images PPaclik RDuin GVan Kempen RKohlus Proceedings of the 16th International Conference on Pattern Recognition the 16th International Conference on Pattern RecognitionQuebec City 2002 * Multivariate texturebased segmentation of remotely sensed imagery for extraction of objects and their uncertainty ALucieer AStein PFisher International Journal of Remote Sensing 26 14 2005 * Texture Classification by Simple Patterns on Edge Direction Movements EswaraReddy B ARao ASuresh VVijaya Kumar IJCSNS International Journal of Computer Science and Network Security 7 11 November 2007 * Face Recognition with Local Binary Patterns TAhonen AHadid MPietikainen Computer Vision, ECCV Proceedings 2004 * Face Recognition Based on the Appearance of Local Regions TAhonen MPietikainen AHadid TMaenpaa 17th International Conference on Pattern Recognition III 2004 * A Coarse-to-Fine Classification Scheme for Facial Expression Recognition XFeng AHadid MPietikainen ICIAR 2004 Proceedings Lecture Notes in Computer Science 3212 II 2004 Image Analysis and Recognition * Facial Expression Recognition with Local Binary Patterns and Linear Programming XFeng MPietikainen AHadid Pattern Recognition and Image Analysis 15 2005 * A Discriminative Feature Space for Detecting and Recognizing Faces AHadid MPietikainen TAhonen IEEE Conference on Computer Vision and Pattern Recognition II 2004 * Texture analysis using gray level run lengths MMGalloway Comput. Graphics Image Process 4 June 1975 * A new approach for unsupervised segmentation WiselinJiji GGanesan L Applied Soft Computing 10 2010 * Comparative analysis of colour models for colour textures based on feature extraction WiselinJiji GGanesan L Int. Jour. of Soft computing 2 3 2007 * Use of gray value distribution of run lengths for texture analysis AChu CM ASehgal JFGreenleaf Pattern Recognition Letters 11 6 1990 * Image characterizations based on joint gray level-run length distributions VBelur EdwinBDasarathy Holder Pattern Recognition Letters 12 1981. August 1991 * Compound Local Binary Pattern (CLBP) for Facial Expression Recognition FaisalAhmed AS MBari EmamHossain 2011 * Image matching using run-length feature Yung-KaunChan Chin-ChenChang Pattern Recognition Letters 22 2001 * A New Method of Texture Classification using various Wavelet Transforms based on Primitive Patterns VVijaya Kumar US NRaju KChandra Sekaran VVKrishna Vision and Image Processing 2008 8 GVIP * Texture Description using Different Wavelet Transforms Based on Statistical Parameters US NRaju VVijaya Kumar ASuresh MRadhika Mani proceedings of the 2nd WSEAS International Symposium on WAVELETS THEORY & APPLICATIONS in Applied Mathematics, Signal Processing & Modern Science (WAV '08) the 2nd WSEAS International Symposium on WAVELETS THEORY & APPLICATIONS in Applied Mathematics, Signal Processing & Modern Science (WAV '08)Istanbul, Turkey 2008 * Survey-texture analysis an no 1983 LVan Gool PDewaele AOosterlinck Computer Vision, Graphics Image Processing 1985 29