Journal of Geodesy and Geoinformation

UCTEA Chamber of Surveying and Cadastre Engineers

TMMOB Harita ve Kadastro Mühendisleri Odası

Jeodezi ve Jeoinformasyon Dergisi

Vol.2  Iss.1  pp.19-28  May 2013  Journal No.107  Doi: 10.9733/jgg.120913.2 www.hkmodergi.org

Analysis of surface textures of physiographic features extracted from multiscale digital elevation models via grey level co-occurrence matrix
Dinesh Sathyamoorthy*
Science & Technology Research Institute for Defence (STRIDE), Ministry of Defence, Malaysia

Abstract
Volume: 2 Issue: 1 Page: 19 - 28 May 2013

This paper is aimed at employing grey level co-occurrence matrix (GLCM) to analyse the surface textures of physiographic features extracted from multiscale digital elevation models (DEMs). Four GLCM parameters, energy, contrast, autocorrelation and entropy, are computed for horizontal (0°), vertical (90°) and diagonal (45 and 135°) cell pair orientations. For the respective DEMs and physiographic features, varying patterns are observed in the plots of the GLCM parameters due to varying surface profiles and the changes that occur over the scales. Due to the smoothing of the terrain during multiscaling, the features have increasing values of energy and entropy, and decreasing values of contrast and entropy, indicating decreasing roughness. Mountains have the highest roughness as compared to the other features over the scales, while basins have the lowest roughness. For each parameter, similar trends are observed in the plots for the four different cell pair orientations, indicating similar trends of change of surface texture in the different orientations over the scales. However, varying values are observed for the different orientations, depending on textural uniformity in the corresponding orientations. The results obtained demonstrate that GLCM can be an appropriate tool for classifying landforms from multiscale DEMs based on the different texture characteristics of the landforms. Keywords Multiscale digital elevation models (DEMs), Physiographic features, Surface texture, Grey level cooccurrence matrix (GLCM), cell pairs.

Özet
Çoklu ölçekli sayısal yükseklik modellerinden çıkarılan fizyografik detaylara ait yüzey doku özelliklerinin gri düzey eş-oluşum matrisi ile analizi
Cilt: 2 Sayı: 1 Sayfa: 19 - 28 Mayıs 2013

Bu çalışmanın amacı çoklu ölçekli Sayısal Yükseklik Modellerinden (SYMler) çıkarılan fizyografik detaylara ait yüzey doku özelliklerinin analizinde Gri Düzey Eş-Oluşum Matrisi (GDEM)’nin kullanılmasıdır. Enerji, kontrast, otokorelasyon ve entropi olmak üzere dört GDEM parametresi yatay (0°), düşey, ve köşegen (45 and 135°) yönler boyunca hücre çiftleri için hesaplanmıştır. Çeşitli yüzey profilleri ve farklı ölçeklerde oluşan değişimler nedeniyle fizyografik özellikler ve bunlara karşılık gelen SYMler için GDEM parametrelerinin çiziminde çeşitli örüntüler gözlemlenmektedir. Çoklu ölçeklendirme esnasında yeryüzündeki detayların yumuşatılması nedeniyle azalan engebeliliği gösterecek şekilde yüzey özellikleri artan enerji ve entropi değerlerine sahip olurken, azalan kontrast ve entropi değerleri oluşmaktadır. Farklı ölçeklerde farklı yüzey özellikleri ile karşılaştırıldıklarında dağlar en yüksek, havzalar ise en düşük engebelilik değerlerine sahip olmaktadır. Her bir parametre için dört farklı hücre çifti yönüne ait çizimlerde benzer eğilimler gözlenmektedir. Yani, farklı ölçeklerde, farklı yönlerde yüzey doku özelliklerindeki değişimde benzer eğilimler oluşmaktadır. Fakat, her bir yöndeki dokusal tekdüzeliğe bağlı olarak farklı yönler için değişen değerler gözlenmiştir. Elde edilen sonuçlar göstermektedir ki, GDEM, yeryüzü şekillerine ait farklı doku özelliklerine dayanan çoklu ölçekli SYMleri kullanarak sınıflandırma yapmak için uygun bir araçtır. Anahtar Sözcükler Çoklu ölçekli sayısal yükseklik modeli (SYM), Fizyografik detaylar, Yüzey doku özelliği, Gri düzey eşoluşum matrisi (GDEM), Hücre çifti.

*Corresponding Author: Tel: +603-8732 4431 Fax: +603-8734 8695 E-mail: dinesh.sathyamoorthy@stride.gov.my

© 2013 HKMO

20 Analysis of surface textures of physiographic features extracted from multiscale digital elevation models via grey level co-occurrence matrix

1. Introduction
Scale variations can constrain the detail with which information can be observed, represented and analysed. The term scale refers to a combination of both spatial extent, and spatial detail or resolution (Tate and Wood 2001; Li 2008; Drăguţ and Eisank 2011; Goodchild 2011). Changing the scale without first understanding the effects of such an action can result in the representation of patterns or processes that are different from those intended due to loss of detail, and variations in terrain parameters and landforms (Lam et al. 2004; Summerfield 2005; Drăgut and Eisank 2011; Goodchild 2011). Hence, feature detection and characterisation often need to be performed at different scales of measurement. Wu et al. (2008), Wood (2009), Drăguţ et al. (2009) and Poulos et al. (2012) demonstrated that analysis of a location at multiple scales allows for a greater amount of information to be extracted from a digital elevation model (DEM) about the spatial characteristics of a feature. A number of studies have been conducted on the classification of various landforms extracted from multiscale DEMs (Fisher et al. 2004; Tay et al. 2005; Schmidt and Andrew 2005; Behrens et al. 2010; Dinesh et al. 2011; Zhao et al. 2012; Chiao et al., 2013). However, not much attention has been provided on the effect of multiscaling on surface textures. Texture refers to the spatial organisation of intensity values of blocks of contiguous cells on a surface. It can be evaluated using perceptual descriptors such as smooth, rough or regular, or regular repetitions of elements or patterns (Haralick et al. 1973; Karu et al. 1996; Srinivasan and Shobha 2008; Gonzalez et al. 2009). This paper is aimed at analysing the surface textures of physiographic features extracted from multiscale DEMs. Physiography (also known as land surface characteristics) is the study of the physical features and attributes of the earth’s land surface. The detection of the physiographic features of a terrain is the first phase involved in the classification of the various landforms of the terrain. A terrain can be segmented into three predominant physiographic features; mountains, piedmont slopes and basins (Miliaresis and Argialas 1999; Dinesh et al. 2007). While a number of methods have been proposed for texture analysis, including Laws’ texture energy (Laws 1980), Markov random field (Cross and Jain 1983), fractal analysis (Pentland 1984), Gabor filtering (Jain and Farrokhnia 1991), wavelet transforms (Chang and Kuo 1993), morphological filtering (Li et al. 1997), quadrature mirror filtering (QMF) (Randen and Husøy 1999a), grey level co-occurrence matrix (GLCM) (Haralick et al. 1973) is employed in this study as it has good performance in terms of spatial localisation (Ohanian and Dubes 1992; Randen and Husøy 1999b; Sharma and Singh 2001; Srinivasan and Shobha 2008). A GLCM is a square matrix with the same size as the number grey levels in an image, which is generated by computing the frequency of occurrences of grey level pairs for cells separated by a fixed geometric displacement. GLCMs have been extensively employed for texture analysis in vari-

ous applications (Baraldi and Parmiggiani 1995; Al-Janobi 2001; Ruiz et al. 2002; Gadelmawla, 2004; Yang et al. 2011; Eichkitz et al. 2013).

2. Methodology 2.1 Data set
The DEM in Figure 1 shows the area of Great Basin, Nevada, USA, which is bounded by latitude 38° 15’ to 42° N and longitude 118° 30’ to 115° 30’W. The DEM was resampled to 925 m in both x and y directions. It is a Global Digital Elevation Model (GTOPO30) and was downloaded from the USGS GTOPO30 website (GTOPO30 1996). GTOPO30 DEMs are available at a global scale, providing a digital representation of the Earth’s surface at a 30 arc-seconds sampling interval. The land data used to derive GTOPO30 DEMs are obtained from digital terrain elevation data (DTED), 1-degree DEMs for USA and the digital chart of the world (DCW). The accuracy of GTOPO30 DEMs varies by location according to the source data. The DTED and the 1-degree dataset have a vertical accuracy of ±30 m while the absolute accuracy of the DCW vector dataset is ±2,000 m horizontal error and ±650 m vertical error (Miliaresis and Argialas 2002). The DEM shows a mountainous terrain with rough surface profile, with tensional forces on the terrain’s crust and thins by normal faulting causing the formation an array of tipped mountain blocks that are separated from broad plain basins, producing a basin-and-range physiography (Howell 1995; Summerfield 1996, 2000; Miliaresis and Argialas 1999; Miliaresis 2008).

2.2 Generation of multiscale DEMs
In this paper, multiscaling is performed using the lifting sche-

Figure 1: The GTOPO30 DEM of Great Basin. The elevation values of the terrain (1,005 to 3,651 m) are rescaled to the interval of 0 to 255 (the brightest cell has the highest elevation). The scale is approximately 1:3,900,000.

Dinesh Sathyamoorthy / Vol.2  Iss.1  2013 21

me (Sweldens 1996, 1997), which has proven to be a powerful multiscale analysis tool in image and signal processing (Claypoole and Baraniuk 2000, Starck 2002, Guo et al. 2008; Abdul-Rahman et al., 2013), and has received recent attention in geospatial analysis (Hayat et al. 2008; Bernadin et al. 2008; Yang et al. 2009; Dinesh et al. 2011; Ahmad Fadzil et al. 2011; Chiao et al. 2013). This is due to its ability to preserve accurate surface profiles, in terms of waveform, shape and amplitude, without causing boundary destruction (Jiang et al. 2001a,b; Nonomura et al. 2010). It is used to decompose an original dataset into low and high frequency subsets using the following three steps: Step 1: Split The original dataset x[n] is divided into two disjoint subsets, even xe[n] and odd xo[n] indexed points. Step 2: Predict The odd and even subsets are often highly correlated. As this correlation structure is typically local, one subset can be used to predict the other subset. In this case, the even indexed subset is be used to predict the odd indexed subset using the prediction operator P (Equation 9). The difference between the predicted and original odd indexed subsets gives the high frequency subset d[n] (Equation 10). The even indexed subset is left unchanged to become the input for the next step in the transform.
P ( xo [n ]) = 1 (xe [n] + xe [n + 1]) 2

Step 3: Update The update step replaces the even indexed subset with an average by applying the update operator U to d[n] (Equation 11) and adding it to the even indexed subset (Equation 12). This gives the low frequency subset c[n], which is a smoother output that represents a coarse approximation to the original dataset.
U (d [n]) = 1 (d [n − 1] + d [n + 1]) 4

(3) (4)

c[n] = x e [n] + U (d [n])

(1) (2)

d [n ] = xo [n ] − P ( xo [n ])

The above three steps form a lifting stage. Using a DEM as the input, an iteration of the lifting stage generates the complete set of multiscale DEMs cs[n] and the elevation loss caused by the change of scale ds[n]. At each iteration, cs[n] only contains half of the points of the input for the iteration, and hence, the resolution of the generated multiscale DEM is reduced by half. At each iteration, the cells of the DEM that are modified are curvatures regions, while the unmodified cells are planar regions. The iterations are repeated until all curvatures in the DEM are removed, leaving only planar regions. For varying DEMs, the number of iterations required would be dependent on the surface profile; a rougher surface profile would require more iterations, while a smoother surface profile would require fewer iterations. Multiscale DEMs of the Great Basin region are generated by implementing the lifting scheme for scales s of 1 to 20. As shown in Figure 2, as the scale increases, the merging of small regions into the surrounding grey level regions increases, causing removal of fine detail in the DEM. As a result, the generated multiscale DEMs possess lower resolutions at higher degrees of scaling.

(a)

(b)

(c)

(d)

(e)

(f)

Figure 2: Multiscale DEMs generated using scales of (a) 1 (b) 3 (c) 5 (d) 10 (e) 15 (f) 20.

22 Analysis of surface textures of physiographic features extracted from multiscale digital elevation models via grey level co-occurrence matrix

(a)

(b)

(c)

(d) Mountains

(e) Piedmont slopes Basins

(f)

Figure 3: Physiographically segmented terrains of the corresponding multiscale DEMs in Figure 2.

2.3 Extraction of physiographic features
The mountains, basins and piedmont slopes of the generated multiscale DEMs are extracted using the mathematical morphological based physiographic segmentation algorithm proposed in Dinesh et al. (2007). Ultimate erosion is used to extract the peaks and pits of the DEM. Conditional dilation is performed on the peaks and pits of the DEM to obtain the mountain and basin cells respectively. The cells that are not classified as mountain cells or basin cells are assigned as piedmont slope cells. As shown in Figure 3, the merging of small regions into the surrounding grey level regions and removal of fine detail in the DEM cause a reduction in the area of the extracted mountains, and an increase in the area of the extracted basins. In general, the area of the piedmont slopes remains fairly constant, but the pattern of the piedmont slopes changes significantly based on the change in pattern of the mountains and basins.

vector quantisation (Ohanian and Dubes 1992; Randen and Husøy 1999b; Aria et al. 2004). For extracting co-occurrence features in this study, the elevation ranges in the DEMs have been reduced from the original ranges to 32 using histogram transformation (Gonzalez et al. 2009). In Cartesian coordinates, GLCM N(i, j) is computed for cell pairs separated by a vector displacement (∆x, ∆y):
N (i, j ) =# {i ( x, y ), j ( x + ∆x, y + ∆y )}

(5)

2.4 GLCM generation
DEMs are 2D matrix arrays with the value of each cell representing elevation at the corresponding location. Computation of co-occurrence features using the full range of cell elevation values could result in a GLCM with a large number of entries with low values, causing the computed texture parameters to be easily distorted by noise in the image. Hence, it is preferable to transform the original elevation range into a smaller number of possible grey levels using either scalar or

where (x, y) and (x + ∆x, y + ∆y) are the base and neighbour cells of the cell pair respectively, i and j are the grey levels of the base and neighbour cells respectively, and # denotes frequency. For this study, in order to evaluate grey level variations over small areas, displacement size of 1 is used. As suggested by previous studies (Haralick et al. 1973; Ohanian and Dubes 1992; Randen and Husøy 1999b; Gadelmawla 2004; Yang et al. 2011; Eichkitz et al. 2013), cell pairs at orientations of 0, 45, 90 and 135 ° are used, corresponding to vector displacements of (∆x = 1, ∆y = 0), (∆x = 1, ∆y = -1), (∆x = 0, ∆y = -1) and (∆x = -1, ∆y = -1) respectively. This allows for the capture of texture features of horizontal (0°), vertical (90°) and diagonal (45 and 135°) orientations. The probability distribution P(i, j) of grey levels in the image is computed as follows:
P(i, j ) = N (i, j ) M

(6)

Dinesh Sathyamoorthy / Vol.2  Iss.1  2013 23

where M is the number of cells in the region under investigation. While a number of GLCM texture parameters have been proposed (Haralick et al. 1973; Ohanian and Dubes 1992; Baraldi and Parmiggiani 1995; Randen and Husøy 1999b), for this study only four are used. The first two parameters, contrast and autocorrelation, measure local variations in the images, while the last two parameters, energy and entropy, measure regularities of distributions of grey levels in the images (Eichkitz et al. 2013). The computation of these four parameters is as follows:” a) Contrast SC (Equation 7) Contrast is a measure of the amount of local variations present in an image. For an image with smooth surface profile, with low levels of local variation, the entries in the GLCM will be clustered along the diagonal of the array, consequently producing a low value of contrast. On the other hand, a high value of contrast implies that the image has a rough surface profile.
S C = ∑∑ (i − j ) • P (i, j )
2 i =0 j =0 3 1 3 1

p y ( j ) = ∑ P (i , j ) i =0

3 1

(10) (11) (12) (13) (14)

µ x = ∑ i • Px (i ) i =0
3 1

3 1

µ y = ∑ j • Py ( j ) j =0

σx =

∑ (i − µ i =0

3 1

x

) 2 • Px (i )

σx =

∑( j − µ j =0

3 1

y

) 2 • Py ( j )

c) Energy SN (Equation 15) Energy provides a measure of repeated transitions of the cell pairs, which indicates textural uniformity. High values of energy occur when the grey level distribution in the image has either a constant or periodic form, with the vector displacements often falling on the same (i, j) grey level pair.

(7)

S N = ∑∑ P 2 (i, j ) i =0 j =0

3 1

3 1

(15)

b) Autocorrelation SR (Equation 8) Autocorrelation is a measure of grey level linear dependencies in the image. A high value of autocorrelation implies a linear relationship between the grey levels of cell pairs, and hence, a smooth surface profile.
SR =

d) Entropy SE (Equation 16)

Entropy is a measure of randomness of the image. If the image is not texturally uniform, many elements in the GLCM will have low values and hence, the entropy will be large.
S E = ∑∑ P (i, j ) • log P (i, j ) i =0 j =0 3 1 3 1

∑∑ i • j • P(i, j ) − µ µ i =0 j =0 x

3 1

3 1

y

(8)

(16)

σ xσ y

3. Results and discussion
The computed GLCM parameters of the generated multiscale DEMs and corresponding extracted physiographic features are shown in Figures 4-7. For the respective DEMs and physiographic features, varying patterns are observed in the plots due to varying surface profiles and the changes that

where μx, μy, σx and σx are the means and standard deviations of px and py (Equations 9-14). p x (i ) = ∑ P (i , j ) j =0 3 1

(9)

Contrast SC

(a) (a) (a) (a)

Contrast SC

(b) (b)

(a)

(b)

(b) (b)

Contrast SC

(c) (c)

(c)

Contrast SC

(d)

(d) (d)

Figure 4: Computed values of contrast for the generated multiscale DEMs and corresponding extracted (c) (d) (c) (d) physiographic 4: Computed pair orientations of (a) 0, (b)the generated multiscale DEMs and Figure Computed values of contrast for 45, (c) 90 and multiscale DEMs and Figure 4:features for cellvalues of contrast for the generated (d) 135°.

corresponding extracted physiographic features for cell pair orientations of (a) corresponding extracted physiographic features for cell pair orientations of (a) 0,0, Figure 4: Computed values of contrast for the generated multiscale DEMs and Figure 4: Computed values of contrast for the generated multiscale DEMs and (b) 45, (c) 90 and (d) 135°. (b) 45, (c) 90 and (d) 135°. corresponding extracted physiographic features for cell pair orientations of (a) 0, corresponding extracted physiographic features for cell pair orientations of (a) 0,

24 Analysis of surface textures of physiographic features extracted from multiscale digital elevation models via grey level co-occurrence matrix

Correlation SR

(a) (a) (a)

Correlation SR

(b) (b) (b)

(a) (a) (a)

(b) (b) (b)

Correlation SR

(c)(c) (c) (c)

Correlation SR

(d) (d) (d) (d)

Figure 5: Computed values of autocorrelation for the generated multiscale DEMs and (c) (d) (c) (d) Figure 5: Computed values of autocorrelationfor the generatedmultiscale DEMs and Figure Computed values of autocorrelation for the for the multiscale multiscale DEMs and Figure5: 5:Computed values of autocorrelation generatedgenerated DEMs and corresponding extracted physiographic features for cell pair orientations of (a) cell 45, (c) 90 and (d) 135°. corresponding extracted physiographic features for0, (b) pair orientations of (a) 0, (b) corresponding extracted physiographic features for cell pair orientations of (a) 0, (b) corresponding extracted physiographic features for cell pair orientations of (a) 0, (b) 45, during gions generated determines the roughness of the surface occur over theFigure 5: Computed valuesthe autocorrelation for the generated multiscale DEMs and scales. The smoothing of of terrain(c) 90 and (d) 135°. a terrain multiscale DEMs and Figure 5: Computed values of autocorrelation for thein 45, (c) 90 and (d) 135°. 45, (c) 90 and (d) 135°. multiscaling causes the removal of curvature regions in the texture. The removal of curvature regions from the terrain corresponding extracted physiographic features for cell pair orientations of (a) (b) corresponding extracted physiographic features for cell pair orientations of (a) 0,0, (b) terrain. Curvature regions, consisting of convex and concave during multiscaling results in the multiscale DEMS and the 45, (c) from corresponding features having increasing values of energy 45, (c) 90 and (d) 135°. crenulations, are used to extract hydrological features 90 and (d) 135°. DEMs, whereby convex and concave crenulations are used and entropy, and decreasing values of contrast and entropy, to extract ridge and drainage networks respectively (Howard indicating decreasing roughness. 1994; Rodríguez-Iturbe and Rinaldo 1997; Sagar et al. 2003; In general, mountains and basins have the highest and Ahmad Fadzil et al. 2011). The distribution of curvature re- lowest distributions of curvature regions respectively. Hen-

Energy SN

(a) (a) (a)

Energy SN

(b) (b) (b)

(a) (a) (a)

(b) (b) (b)

Energy SN

(c)(c) (c) (c)

Energy SN

(d) (d) (d) (d)

corresponding extracted physiographic features for cell pair orientations of (a) 0, corresponding extracted physiographic features for cell pair orientations of (a) 0, 45, (c) 90 and (d) 135°. 45, (c) 90 and (d) 135°.

corresponding extracted physiographic features for cell pair orientations of (a) 0, (b) (b) corresponding extracted physiographic features for cell pair orientations of (a) 0, 45,energyfor the generated multiscale DEMs and (c) 90 and the 135°. (d) generated multiscale DEMs and Figure 6: Computed values of energy and (d) 135°. Figure 6: Computed values 45,(c) 90 and (d) 135°. of 45, (c) 90 for (b) (b)

(d) (c) (d) FigureFigure6: Computed values of energy for the generated multiscale DEMs and 6: Computed(c) values of values of energy for multiscale DEMs and corresponding and Figure 6: Computed energy for the generatedthe generated multiscale DEMs extracted physiographic features for cell physiographic features 45, (c) 90pair(d) 135°. pair orientations of (a) 0, (b) for cell and orientations of (a) 0, corresponding extracted (b)

Figure 6: Computed values of energy for the generated multiscale DEMs and

(a)

Dinesh Sathyamoorthy / Vol.2  Iss.1  2013 25

(b)

Entropy SE

(a) (a) (a)

Entropy SE

(b) (b) (b)

(a) (a) (c)

(b) (b) (d)

Figure 7: Computed values of entropy for the generated multiscale DEMs and
Entropy SE

corresponding extracted physiographic features for cell pair orientations of (a) 0, (b) 45, (c) 90 and (d) 135°.

(c) (c) (c) (c)

Entropy SE

(d) (d) (d) (d)

corresponding extracted physiographic features for cell pair orientations of (a) 0, (b) 45, corresponding extracted physiographic features for cell pair orientations ofof (a) 0, (b) 45, corresponding extracted physiographic features for cell pair orientations (a) 0, (b) 45, ce, it is observed thatFigure 7: Computed values ofhave the forlowest roughness. On the DEMs hand, the lowest values of for all the scales, mountains (c) 90 and (d) 135°. entropy generated multiscale other and (c) 90 and (d)the135°. 135°. (c) 90 and (d) highest roughness as compared to the other features, while energy and entropy, and the highest values of contrast and corresponding extracted physiographic features for cell pair orientations of (a) 0, (b) 45, basins have the lowest roughness. For autocorrelation, the entropy are observed for the diagonal orientations (45 and (c) 90 and 135°), DEMs and mountains exhibit similar patterns at the initial (d) 135°.indicating that the textural features in these directions scales, indicating that for terrains with rough surface profi- have the highest roughness. les, mountains are the more dominant of the three predomiThe results obtained thus far are based on analysis of a nant physiographic features in deciphering terrain character. mountainous terrain with rough surface profile. The analyFor each parameter, similar trends are observed in the sis is further extended for GTOPO30 DEMs of Great Plains, plots for the four different cell pair orientations, indicating Nebraska (bounded by latitude 39° to 43° N and longitude similar trends of change of surface texture in the different 98° to 101° W) with moderate surface profile, and Great Falorientations over the scales. However, varying values are ob- ls, Montana (bounded by latitude 36° to 40° N and longitude served for the different orientations, depending on textural 75° to 79° W) with smooth surface profile (Figure 8). The uniformity in the corresponding orientations. The highest va- smoother surface profiles of these terrains results in fewer lues of energy and entropy, and the lowest values of contrast multiscaling iterations required to remove all the curvature and entropy are observed for the vertical orientation (90°), regions; 11 and 4 for the moderate and smooth surface proindicating that the textural features in this direction have the files respectively. The smoother surface profiles also result

Figure 7: Computed values entropy for the generated multiscale DEMs extracted (d) FigureFigure 7: Computed entropy ofentropy for the generated multiscale DEMs and 7: Computed(c) values values ofof entropy for multiscale DEMs and corresponding and Figure 7: Computedof values for the generated the generated multiscale DEMs and physiographic features for cell pair orientations of (a) 0, (b) 45, (c) 90 and (d) 135°.

(a)

(b)

Figure 8: GTOPO30 DEMs of (a) Great Plains (elevation range of 25 to 1,736 m) and (b) Great Falls (elevation range of 0 to 1,105 m). The scale is approximately 1:3,900,000.

26 Analysis of surface textures of physiographic features extracted from multiscale digital elevation models via grey level co-occurrence matrix

(a)

(b)

Contrast SC

(a) (a) (c) (a)

Entropy SE

(b) (b)

(a)

(b) (d) (b)

Figure 9: Computed values of (a) contrast, (b) autocorrelation, (c) energy and (d) entropy for the generated multiscale DEMs of Great Plains and corresponding extracted physiographic features for cell pair orientation of 0°.
Entropy SE Energy SN

(c) (c) (c)

(d)(d) (d)

(d) entropy for the generated multiscale DEMs of Great Plains and corresponding 9: Computed and decreasing for cell pair orientation of 0°. in increasing values ofFigureextracted physiographic features landforms. Surface texture0°. often used in conjunction energy and entropy,values of (a) contrast, (b) autocorrelation, (c) energy and extracted physiographic features for cell pair orientation of has values of contrast and entropy (Figures 9 and 10), indica- DEMs of Great data for image classification (Woodcock and (d) entropy for the generated multiscale with spectral Plains and corresponding ting decreasing roughness. Similar to the terrain with rough extracted physiographic featuresStrahler pair orientation Miranda 1998; Randen and Husøy for cell 1987; Carr and of 0°. surface profile, mountains have the highest roughness, while 1999a; van Ginneken and Haar Romeny 2003; Yang et al. basins have the lowest roughness. The results obtained demonstrate that GLCM can be an 2011). The use of texture implies that landforms in the DEM appropriate tool for classifying landforms from multiscale are not spatially homogenous, but rather the inhomogeneity DEMs based on the different texture characteristics of the of the landforms produce different texture characteristics. (a) (b)

Figure 9: Computed values of (a) contrast, (a) autocorrelation, (c) energy and (d) entropy for the generated Figure 9: Computed values of (b) contrast, (b) autocorrelation, (d) energy and (c) (c) multiscale DEMs of Great Plains and corresponding extracted physiographic (c) energy and pair Figure 9: Computed values of (a) contrast, (b) autocorrelation, features for cell orientation of 0°.for the generated multiscale DEMs of Great Plains and corresponding (d) entropy

(a) (a) (a) (a) (a) (c)
Entropy SE Energy SN

Correlation SR

Energy SN

(b) (b) (b)(b) (b) (d)

Figure 10: Computed values of (a) contrast, (b) autocorrelation, (c) energy and (d) entropy for the generated multiscale DEMs of Great Falls and corresponding extracted physiographic features for cell pair orientation of 0°. (c) (c) (c) (c)

(d) (d)
(d) (d)

Figure 10: Computed values valuescontrast, (b) autocorrelation, (c) energy and (d) entropy for the Figure 10: Computed of (a) of (a) contrast, (b) autocorrelation,(d) energy and (c) (c) generated multiscale DEMs of Great Falls and corresponding extracted physiographic features for cell pair orientation of 0°. the generated multiscale DEMs of Great Falls and corresponding (d) entropy for

Figure 10: Computed values of (a) contrast, (b) autocorrelation, (c) energy and Figure 10: Computed values of (a) contrast, (b) autocorrelation, (c) energy and Figure extracted physiographic (a) contrast, cell autocorrelation, of 0°. 10: Computed values of features for (b) pair orientation corresponding (d) entropy for the generated multiscale DEMs of Great Falls and(c) energy and (d) entropy for the generated multiscale DEMs of Great Falls and corresponding (d) entropy for the physiographic features for cell Great Falls and corresponding extracted generated multiscale DEMs of pair orientation of 0°. extracted physiographic features for cell pair orientation of 0°. extracted physiographic features for cell pair orientation of 0°.

Dinesh Sathyamoorthy / Vol.2  Iss.1  2013 27

4. Conclusion
In this study, GLCM was used to analyse the surface texture of physiographic features extracted from multiscale DEMs. For the respective DEMs and physiographic features, varying patterns are observed in the plots of GLCM parameters due to varying surface profiles and the changes that occur over the scales. Due to the smoothing of the terrain during multiscaling, the features have increasing values of energy and entropy, and decreasing values of contrast and entropy, indicating decreasing roughness. Mountains have the highest roughness as compared to the other features over the scales, while basins have the lowest roughness. For autocorrelation, the DEMs and mountains exhibit similar patterns at the initial scales, indicating that mountains are the more dominant of the three predominant physiographic features in deciphering terrain character. For each parameter, similar trends are observed in the plots for the four different cell pair orientations, indicating similar trends of change of surface texture in the different orientations over the scales. However, varying values are observed for the different orientations, depending on textural uniformity in the corresponding orientations. The results obtained demonstrate that GLCM can be an appropriate tool for classifying landforms from multiscale DEMs based on the different texture characteristics of the landforms.

References
Abdul-Rahman H.S., Jiang X.J., Scott P.J., (2013), Freeform surface filtering using the lifting wavelet transform, Precision Engineering, 37(1), 187-202, doi: 10.1016/j. precisioneng.2012.08.002. Ahmad Fadzil M.H., Dinesh S., Vijanth Sagayan A., (2011), A method for computation of surface roughness of digital elevation model terrains via multiscale analysis, Computers & Geosciences, 37(2), 177-192, doi: 10.1016/j.cageo.2010.05.021. Al-Janobi A., (2001), Performance evaluation of cross-diagonal texture matrix method of texture analysis, Pattern Recognition, 34(1), 187-197, doi: 10.1016/S0031-3203(99)00206-X. Aria E.H., Saradjian M.R., Amini J., Lucas C., (2004), Generalized cooccurrence matrix to classify IRS-1D images using neural network, XXth ISPRS Congress, 12-23 July 2004 Istanbul, Turkey. Baraldi A., Parmiggiani F., (1995), An investigation of the textural characteristics associated with gray level co-occurrence matrix statistical parameters, IEEE Transactions on Geoscience and Remote Sensing, 33(2), 293-302, doi: 10.1109/36.377929. Behrens T., Zhu A.X., Schmidt K., Scholten T., (2010), Multiscale digital terrain analysis and feature selection for digital soil mapping, Geoderma, 155(3-4), 175-185, doi: 10.1016/j. geoderma.2009.07.010. Bernardin T., Cowgill E. S., Gold R. D., Hamann, B., Kreylos O., Schmitt A., (2008), Real-time terrain mapping, In: Scientific Visualization: Challenges for the Future, (Hagen H., Ed.), IEEE Computer Society Press, Los Alamitos, California, pp. 275-288. Carr J.R., Miranda F.P., (1998), The semivariogram in comparison to the co-occurance matrix for classification of image texture, IEEE Transactions on Geosciences and Remote Sensing, 36(6): 1945-1952, doi: 10.1109/36.729366. Chang T., Kuo C.C.J., (1993), Texture analysis and classification with tree-structured wavelet transform, IEEE Transactions in Image Processing, 2(4), 429-441, doi: 10.1109/83.242353. Chiao L.Y., Hsieh C., Chiu T.S., (2013), Exploring spatiotemporal

ecological variations by the multiscale interpolation, Ecological Modelling, 246(10), 26-33. Cross G.R., Jain A.K., (1983), Markov random field texture models, IEEE Transactions on Pattern Analysis and Machine Intelligence, 5(1), 25-39, doi: 10.1109/TPAMI.1983.4767341. Claypoole R.L., Baraniuk R.G., (2000), A multiresolution wedgelet transform for image processing, In Wavelet Applications in Signal and Image Processing VIII, Volume 4119 of SPIE Proceedings (Unser M.A., Aldroubi A., Laine A.F., Eds.), SPIE, Bellingham, Washington, pp. 253-262. Dinesh S., Radhakrishnan P., Sagar B.S.D., (2007), Morphological segmentation of physiographic features from DEM, International Journal of Remote Sensing, 28(15), 3379-3394, doi: 10.1080/01431160500486708. Dinesh S., Ahmad Fadzil M.H., Vijanth Sagayan A., (2011), Computation of uncertainty of physiographic features extracted from multiscale digital elevation models using fuzzy classification, Image Processing, Image Analysis and RealTime Imaging (IPIARTI) Symposium 2011, 6 October 2011, Multimedia University (MMU), Cyberjaya. Drăguţ L., Eisank C., (2011), Object representations at multiple scales from digital elevation models, Geomorphology, 129 (34), 183-189, doi: 10.1016/j.geomorph.2011.03.003. Drăguţ L., Schauppenlehner T., Muhar A., Strobl J., Blaschke T., (2009), Optimization of scale and parametrization for terrain segmentation: An application to soil-landscape modeling, Computers & Geosciences, 35(9), 1875-1883, doi: 10.1016/j. cageo.2008.10.008. Eichkitz C.G., Amtmann J. and Schreilechner M.G (2013), Calculation of grey level co-occurrence matrix-based seismic attributes in three dimensions, Computer & Geosciences, 30, 176-183, doi: 10.1016/j.cageo.2013.07.006. Fisher P., Wood J., Cheng T., (2004), Where is Helvellyn? Multiscale morphometry and the mountains of the English Lake District, Transactions of the Institute of British Geographers, 29, 106128, doi: 10.1111/j.0020-2754.2004.00117.x. Gadelmawla E.S., (2004), A vision system for surface roughness characterisation using gray-level co-occurrence matrix, NDT&E International, 37(7), 577-588, doi: 10.1016/j. ndteint.2004.03.004. Gonzalez R.C., Woods R.E., Eddins S.I., (2009), Digital Image Processing Using MATLAB, Prentice Hall, New Jersey. Goodchild M.F., (2011), Scale in GIS: An overview, Geomorphology, 130(1-2), 5-9, doi: 10.1016/j.geomorph.2010.10.004. GTOPO30, (1996), GTOPO30: Global 30 Arc Second Elevation Data Set, Available online at: http://edcwww.cr.usgs.gov/ landdaac/gtopo30/gtopo30.html, [ Accessed 1 June 2009]. Guo S.M., Chang W.H., Tsai J.S.H., Zhuang B.L., Chen L.C., (2008), JPEG 2000 wavelet next term filter design framework with chaos evolutionary programming, Signal Processing, 88(10): 2542-2553, doi: 10.1016/j.sigpro.2008.05.009. Haralick R.M., Shanmugam K., Dinstein I., (1973), Texture features for image classification, IEEE Transactions on Systems, Man and Cybernetics, 3(6), 610-621, doi: 10.1109/ TSMC.1973.4309314. Hayat K, Puech W., Gesquère G., (2008), Scalable 3-D terrain visualization through reversible JPEG2000-based blind data hiding, IEEE Transactions on Multimedia, 10(7), 1261-1276, doi: 10.1109/TMM.2008.2004905. Howard A.D., (1994), A detachment-limited model of drainage basin evolution, Water Resources Research, 30(7), 2261-2285, doi: 10.1029/94WR00757.

28 Analysis of surface textures of physiographic features extracted from multiscale digital elevation models via grey level co-occurrence matrix

Howell D., (1995), Principles of Terrain Analysis: New Applications for Global Tectonics, Chapman and Hall, London. Jain A.K., Farrokhnia F., (1991), Unsupervised texture segmentation using Gabor filters, Pattern Recognition, 24(12), 1167-1186, doi: 10.1016/0031-3203(91)90143-S. Jiang X.Q., Blunt L., Stout K.J., (2001a), Application of the lifting scheme to rough surfaces, Precision Engineering, 25(2), 83-89, doi:10.1016/S0141-6359(00)00054-4. Jiang X.Q., Blunt L., Stout K.J., (2001b), Lifting wavelet for threedimensional surface analysis, International Journal of Machine Tools and Manufacture, 41(13-14), 2163-2169, doi: 10.1016/ S0890-6955(01)00083-9. Karu K., Jain A.K., Bolle R.M., (1996), Is there any texture in the image?, Pattern Recognition, 29(9), 1437-1446, doi: 10.1016/0031-3203(96)00004-0. Lam N., Catts D., Quattrochi D., Brown D., McMaster R., (2004), Scale, In: A Research Agenda for Geographic Information Science, (Rechcígl M., McMaster R.B., Usery E.L., Eds.), CRC Press, New York, pp. 93-128. Laws K.I., (1980), Textured Image Segmentation, PhD Thesis, University of Southern California, Los Angeles, California. Li Z., (2008), Multi-scale digital terrain modelling and analysis, In Advances in Digital Terrain Analysis (Zhou Q., Lees B., Tang G., Eds.), Springer, Berlin, pp. 59-83. Li W., Haese-Coat V., Ronsin J., (1997), Residues of morphological filtering by reconstruction for texture classification, Pattern Recognition, 30(7), 1081-1093, doi: 10.1016/S00313203(96)00146-X. Miliaresis G. (2008). Quantification of terrain processes, Lecture Notes in Geoinformation and Cartography, XIV: 13-28. Miliaresis G., Argialas D.P., (1999), Segmentation of physiographic features from Global Digital Elevation Model/GTOPO30, Computers & Geosciences, 25(7), 715-728, doi: 10.1016/ S0098-3004(99)00025-4. Miliaresis G., Argialas D.P., (2002), Quantitative representation of mountain objects extracted from the Global Digital Elevation Model (GTOPO30), International Journal of Remote Sensing, 23(5), 949-964, doi: 10.1080/01431160110070690 Nonomura K., Ono M., Zhou L.B., Shimizu J., Ojima H., (2010), Design of digital filters for Si wafer surface profile measurement: Noise reduction by lifting scheme wavelet transform, Advanced Materials Research, 126-128, 732-737., doi: 10.4028/www. scientific.net/AMR.126-128.732. Ohanian P.P., Dubes R.C., (1992), Performance evaluation for four classes of textural features, Pattern Recognition, 25(8), 819833, doi: 10.1016/0031-3203(92)90036-I. Pentland A., (1984), Fractal-based description of natural scenes, IEEE Transactions on Pattern Analysis and Machine Intelligence, 6(6), 661-674, doi: 10.1109/TPAMI.1984.4767591. Poulos M.J., Pierce J.L., Flores A.N., Benner S.G., (2012), Hillslope asymmetry maps reveal widespread, multi-scale organization, Geophysical Research. Letters, 39(6), L06406, doi: 10.1029/2012GL051283. Randen T., Husøy J.H., (1999a), Texture segmentation using filters with optimized energy separation, IEEE Transactions on Image Processing, 8(4), 571-582, doi: 10.1109/83.753744. Randen T., Husøy J.H., (1999b), Filtering for texture classification: A comparative study, IEEE Transactions on Pattern Analysis and Machine Intelligence, 21(4), 291-310, doi: 10.1109/34.761261. Rodríguez-Iturbe I., Rinaldo A., (1997), Fractal River Basins: Chance and Self-Organization, Cambridge University Press, New York. Ruiz L.A., Sarría A.F., Recio J., (2002), Evaluation of texture analysis techniques to characterize vegetation, In: Recent Advances in Quantitative Remote Sensing, (Sobrino J.A., Ed.), Universitat de València, Valencia, pp. 514-521. Sagar B.S.D., Murthy M.B.R., Rao C., Raj B., (2003), Morphological

approach to extract ridge-valley connectivity networks from digital elevation models, International Journal of Remote Sensing, 24(3), 573-581, doi: 10.1080/01431160304983. Schmidt J., Andrew R., (2005), Multi-scale landform characterization, Area, 37(3), 341-350, doi: 10.1111/j.14754762.2005.00638.x. Sharma M., Singh S., (2001), Evaluation of texture methods for image analysis, The Seventh Australian and New Zealand Intelligent Information Systems Conference (ANZIIS 2001), 18-21 November 2001, Perth, Australia. Srinivasan G.N., Shobha G. (2008), Statistical texture analysis, Proceedings of World Academy of Science, Engineering and Technology, 36, 1264-1269. Starck J.L., (2002), Non-linear multiscale transforms, In: Multiscale and Multiresolution Methods: Theory and Applications. Springer, (Barth T.J., Chan T., Haimes R., Eds.), Heidelberg, Germany, pp. 239-266, doi: 10.1007/978-3-642-56205-1_5. Summerfield M., (1996), Global Geomorphology, Longman, Essex. Summerfield M., (Ed.), (2000), Geomorphology and Global Tectonics, John Wiley & Sons, New York. Summerfield M., (2005), A tale of two scales, or two geomorphologies, Transactions of the Institute of British Geographers, 30(4), 402415, doi: 10.1111/j.1475-5661.2005.00182.x. Sweldens W., (1996), The lifting scheme: A custom-design construction of biothorgonal wavelets, Applied and Computational Harmonic Analysis, 3(2), 186-200, doi: 10.1006/acha.1996.0015. Sweldens W., (1997), The lifting scheme: A construction of second generation wavelets, Journal on Mathematical Analysis, 29(2), 511-546, doi: 10.1137/S0036141095289051. Tate N., Wood J., (2001), Fractals and scale dependencies in topography, In: Modelling Scale in Geographical Information Science (Tate N., Atkinson P., Eds.), Wiley, Chicester, pp. 35-51. Tay L. T., Sagar B.S.D., Chuah H.T., (2005), Analysis of geophysical networks derived from multiscale digital elevation models: A morphological approach, IEEE Geoscience and Remote Sensing Letters, 2(4), 399-403, doi: 10.1109/ LGRS.2005.856008. van Ginneken B., Haar Romeny B.M., (2003), Multi-scale texture classification from generalised locally orderless images, Pattern Recognition, 36(4), 899-911, doi: 10.1016/S00313203(02)00118-8. Wood J., (2009), Geomorphometry in LandSerf, In: Developments in Soil Science, vol. 33: Geomorphometry: Concepts, Software, Applications, (Hengl T., Reuter H.I., Eds.), Elsevier, Armsterdam, pp. 333-349. Woodcock C.E., Strahler A.H., (1987), The factor of scale in remote sensing, Remote Sensing of Environment, 21(3), 311-332, doi: 10.1016/0034-4257(87)90015-0. Wu S., Li J., Huang G.H., (2008), A study on DEM-derived primary topographic attributes for hydrologic applications: Sensitivity to elevation data resolution, Applied Geography, 28(3), 210223, doi: 10.1016/j.apgeog.2008.02.006. Yang M.D., Su T.C., Pan N..F, Liu P., (2011), Feature extraction of sewer pipe defects using wavelet transform and co-occurrence matrix, International Journal of Wavelets, Multiresolution and Information Processing, 9, 211-225, doi: 10.1142/ S0219691311004055. Yang Z., Liu H., Yang N., Xiao X., (2009), Study on multiscale generalization of DEM based on lifting scheme, In: Proceedings of the SPIE, Volume 7492, (Liu, Y., Tang, X., Eds.), SPIE, Bellingham, Washington, pp. 74922E-74922E-6. Zhao Y., Kazuo K., Fujita F., (2012), Multi-scale decomposition of co-seismic deformation from high resolution DEMs: A case study of the 2004 mid-Niigata earthquake, Acta Geoligica Sinica, 86(4), 1013-1021, doi: 10.1111/j.1755-6724.2012.00725.x.