Gao 2003 Geophysics

Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

GEOPHYSICS, VOL. 68, NO. 4 (JULY-AUGUST 2003); P. 12941302, 9 FIGS., 1 TABLE. 10.1190/1.

1598122

Volume texture extraction for 3D seismic visualization and interpretation

Dengliang Gao
cause of the subtlety of amplitude variations and limitation in data visibility in 3D space, it is difcult for them to extract quantitative information for automatic feature discrimination, visualization, and detection. In previous studies, various seismic attributes have been extracted from the amplitude in an attempt to facilitate seismic feature identication and interpretation. These efforts (e.g., Taner and Sheriff, 1977) have signicantly enhanced interpreters ability to discriminate and visualize geological features efciently and objectively. However, very few attributes (e.g., Taner et al., 1994; Bahorich and Farmer, 1995; M. T. Taner, 1998, personal communication; Marfurt et al., 1999) have been published to recognize certain seismic features, for example, those dened by both intratrace and intertrace relationships of amplitude from a 3D perspective. To overcome these limitations and difculties, I introduce a new approach to the problem by extracting volume seismic textures using a point-relational statistical method. An image texture is a general term that refers to a characteristic pattern dened by the magnitude and variation of neighboring data samples at a given location in a physical space. Although studies of image texture have been published since the 1970s, the early concept was primarily applied to twodimensional (2D) image analysis (e.g., Haralick et al., 1973; Weszka et al., 1976; Reed and Hussong, 1989; Gao et al., 1998). Little has been published on its application to reection seismic data visualization and interpretation (e.g., Zhang and Simaan, 1989; Vinther et al., 1996; Gao, 1999a, b, 2001a, b, 2002). In this paper, I describe a methodology to characterize 3D seismic textures and investigate its potential geological implications. Such a methodology represents a new, effective approach to discriminating and visualizing seismic features that may not be easily recognizable using visual inspection and conventional attribute extraction algorithms.
CONCEPTS AND METHODOLOGY

ABSTRACT

Visual inspection of poststack seismic image patterns is effective in recognizing large-scale seismic features; however, it is not effective in extracting quantitative information to visualize, detect, and map seismic features in an automatic and objective manner. Although conventional seismic attributes have signicantly enhanced interpreters ability to quantify seismic visualization and interpretation, very few attributes are published to characterize both intratrace and intertrace relationships of amplitudes from a three-dimensional (3D) perspective. These relationships are fundamental to the characterization and identication of certain geological features. Here, I present a volume texture extraction method to overcome these limitations. In a two-dimensional (2D) image domain where data samples are visualized by pixels (picture elements), a texture has been typically characterized based on a planar texel (textural element) using a gray level co-occurrence matrix. I extend the concepts to a 3D seismic domain, where reection amplitudes are visualized by voxels (volume picture elements). By evaluating a voxel co-occurrence matrix (VCM) based on a cubic texel at each of the voxel locations, the algorithm extracts a plurality of volume textural attributes that are difcult to obtain using conventional seismic attribute extraction algorithms. Case studies indicate that the VCM texture extraction method helps visualize and detect major structural and stratigraphic features that are fundamental to robust seismic interpretation and successful hydrocarbon exploration.

INTRODUCTION

Since the early 1980s, three-dimensional (3D) seismic imaging technology has signicantly contributed to subsurface geologic mapping and hydrocarbon exploration in the petroleum industry. From high-quality 3D seismic data, exploration geologists are able to recognize large-scale seismic features by visual inspection of seismic reection patterns. However, be-

A seismic texture, as opposed to other image textures, is dened as a reection amplitude pattern that is characterized by the magnitude and variation of neighboring acoustic samples at a given location in a seismic volume (Gao, 1999a, b, 2001a, b, 2002). At each of the sample locations, a seismic texture is evaluated by analyzing an array of neighboring reection

Manuscript received by the Editor December 26, 2001; revised manuscript received February 3, 2003. Marathon Oil Corporation, Computer-Aided Interpretation, P.O. Box 3128, Houston, Texas 77253-3128. E-mail: [email protected]. c 2003 Society of Exploration Geophysicists. All rights reserved. 1294

Volume Seismic Texture Analysis

1295

amplitudes. Such an array of reection amplitudes is here referred to as a seismic texture element (texel) (Figure 1) (e.g., Haralick et al., 1973; Reed and Hussong, 1989; Gao et al., 1998; Gao, 1999a, b, 2001a, b, 2002), which is geometrically equivalent to the analysis window commonly used in various seismic processing and attribute extraction algorithms. In previous 2D image texture analysis, a texel is typically a rectangular or square that is formed by a nite number of neighboring pixels (picture elements). In the 3D seismic domain, a texel is a minicube that consists of N x N y Nz voxels (volume picture elements) in the inline, crossline, and vertical directions, respectively (Figures 1 and 2a). The texel size and aspect ratio are exible and dependent upon the exploration objectives. Typically, N x and N y range from 3 to 9, and Nz ranges from 7 to 21 to extract meaningful textural information. In certain cases, however, the texel size and aspect ratio can be quite different to achieve special objectives. For example, a horizontal window ( N x = N y Nz ), which is equivalent to that commonly used in horizontal image analysis, can be used to emphasize the lateral but not vertical variations in amplitude. A vertical window ( N y = Nz N x or N x = Nz N y ), which is equivalent to that used in line-based seismic interpretation, captures lateral variations in the inline or crossline direction, but not both. A trace segment ( N x = N y Nz ), which is equivalent to that commonly used in trace-based attribute analysis, is typically used to characterize waveform but not the trace-to-trace variations. Fundamentally different from other digital images, a 3D reection seismic image consists of vertical traces with alternating positive and negative amplitudes (Figure 1). These amplitudes of opposite polarities are aligned laterally in both inline

and crossline directions to form a coherent stratal pattern of reection amplitudes. To characterize such a unique pattern, I choose to evaluate the point-relational statistics on a 3D texel basis using a voxel co-occurrence matrix (VCM) (Gao, 1999b) that is equivalent to the gray level co-occurrence matrix (GLCM) previously used in 2D image texture analysis (e.g., Haralick et al., 1973; Reed and Hussong, 1989; Gao et al., 1998). The VCM is a statistical representation of the amplitude pattern of a texel in a tabular format. More specically, if a seismic data set has N g gray levels ( N g = 256 for 8-bit data), the VCM for a texel at each of the sample locations is a square symmetrical matrix consisting of N g N g elements. By denition, the element E (i , j , , ) at i th row and j th column of the matrix denotes the number of times (frequency) in the texel that a voxel with amplitude i (< Ng ) is neighbored by a voxel with amplitude j (< N g ) in the direction of and (Figures 2a and 2b). Here, and denote the azimuth and dip of a vector, respectively (Figure 2b), along which the voxel co-occurrence is evaluated. Due to the stratal pattern of the seismic images, the VCM is normally different in different directions. For three orthogonal directions (Figure 2b) along the x -axis ( = 0 , = 0 ), y -axis ( = 90 , = 0 ), and z -axis ( = 90 ), for example, the elements E (i , j , , ) of the respective VCM can be mathematically expressed as follows (Reed and Hussong, 1989):

E (i , j , 0, 0) =

{((m , n , o), ( p , q , r ) (x , y , z )), (|m p | = 1, n q = 0, o r = 0, g (m , n , o) = i , g ( p , q , r ) = j )}, (1)

FIG. 1. Four cubic texel (3D texture element) examples at four different locations in a seismic amplitude volume. A cubic texel, which consists of a 3D array of spatially associated voxels (volume picture element) at each of the sample locations, is fundamental for 3D image feature discrimination and visualization. At the shallow structural level, for example, there are differences in internal textures between the high-amplitude laterally coherent interval (A) and the low-amplitude discontinuous interval (B). At the deep structural level, the inclined reection pattern (C) on the hanging wall is distinct from the reection pattern on the footwall (D) of the listric fault. Such different reection patterns can be identied, visualized, and mapped quantitatively using the texture extraction method (see Figures 5 and 8).

1296

Gao

E (i , j , 90, 0) =

{((m , n , o), ( p , q , r ) (x , y , z )), (m p = 0, |n q | = 1, o r = 0, g (m , n , o) = i , g ( p , q , r ) = j )}, (2)

E (i , j , , 90) =

{((m , n , o), ( p , q , r ) (x , y , z )), (m p = 0, n q = 0, |o r | = 1, g (m , n , o) = i , g ( p , q , r ) = j )}, (3)

where denotes the total number of times that the voxel cooccurrence relationship dened in the braces exists in the texel, (x , y , z ) represents the volume extent of the 3D seismic image,

and g (m , n , o) and g ( p , q , r ) stand for the values of the two voxels at (m , n , o) and ( p , q , r ) in a texel, respectively. From the VCM, a plurality of textural attributes are derived, each of which describes a specic textural feature of the texel. Based on the comparison, I found that texture homogeneity, contrast, and randomness are among the most effective ones in characterizing seismic data. Texture homogeneity highlights the overall smoothness of amplitude and texture contrast emphasizes the magnitude of differences in amplitude of neighboring voxels, whereas texture randomness measures the amplitude predictability from one voxel to the next. Although there is a certain degree of correlation among these three textural attributes, the correlation is nonlinear, and each attribute should contribute to minimizing the nonuniqueness in texture discrimination. These three textural attributes are computed using the following equations (Haralick et al., 1973; Reed and

START Retrieve seismic amplitude data Select texel size and geometry Select texture orientation Select first/next voxel location Build and requantize texel Build voxel cooccurrence matrix (VCM) Calculate texture attributes Rescale and store attributes to attribute volumes
Y

Next voxel location ?


FIG. 2. (a) A schematic representation of a typical seismic cubic texel (3D). The texel can also be planar (2D) and linear (1D). The digits in the texel denote the 4-bit amplitude (16 intensity levels) requantized from the original 8-bit (256 intensity levels) input data. The requantization is performed to enhance the computational efciency (see discussion in the text). (b) A schematic notation dening the direction in which the point-relational (voxel co-occurrence) statistics are evaluated. Typically, the point-relational statistics are evaluated along the inline-horizontal ( = 0 , = 0 ), crossline-horizontal ( = 90 , = 0 ), and vertical ( = 90 ) directions, respectively.

Visualize and interpret attribute volumes


FIG. 3. A workow chart for VCM seismic texture analysis. The input is a single amplitude volume (e.g., Figure 1). After texture extraction by evaluating textural attributes at each voxel location along different directions (Figure 2), the algorithm outputs a plurality of texture attribute volumes for subsequent interpretation (e.g., Figures 49).

Volume Seismic Texture Analysis

1297

Hussong, 1989):
n

Homogeneity =
i =1 n 1

[ E (i , j , , )/R ]2 ,
n n

(4)

Contrast =

m2
m =0 n i =1 j =1 |i j |=m

E (i , j , a , )/R ,

(5)

Randomness =
i =1

E (i , j , a , )/R log[ E (i , j , a , )/R ], (6)

where E (i , j ) represents the element at the i th row and the j th column of the VCM, and n is the dimension of the VCM. R is a normalization constant representing the maximum possible times of the co-occurrence. Along the x (inline), y (crossline), and z (time or depth) directions, for example, R is dened by Rx = 2( N x 1) N y Nz , R y = 2 N x ( N y 1) Nz , and Rz = 2 N x N y ( Nz 1), respectively. To demonstrate the procedure, examine a texel (Figure 2a) consisting of 9 5 9 voxels that are requantized to 16 gray levels (4-bit precision) from the original 256 gray levels (8-bit precision). The following matrices (VCMx , VCMy , and VCMz ) are three VCMs that are derived from the texel along the x , y , and z directions, respectively.

0 20 5 5 0 0 0 0 VCMx = 0 0 0 0 0 0 0 0 24 0 0 0 0 0 0 0 VCMy = 0 0 0 0 0 0 0 0

20 5 0 10 10 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 24 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 8 0 0 0 0 0 0 0 0 0 0 0 0

0 0 0 0 80 15 10 5 0 0 0 0 0 0 0 0 0 0 0 0 96 0 0 0 0 0 0 0 0 0 0 0

0 0 0 0 15 10 35 0 0 0 0 0 0 0 0 0 0 0 0 0 0 48 0 0 0 0 0 0 0 0 0 0

0 0 0 0 10 35 10 25 0 0 0 0 0 0 0 0 0 0 0 0 0 0 80 0 0 0 0 0 0 0 0 0

0 0 0 0 5 0 25 10 15 10 0 0 0 0 0 0 0 0 0 0 0 0 0 56 0 0 0 0 0 0 0 0

0 0 0 0 0 0 0 15 0 30 0 0 0 0 0 0 0 0 0 0 0 0 0 0 48 0 0 0 0 0 0 0

0 0 0 0 0 0 0 10 30 0 5 5 0 0 0 0 0 0 0 0 0 0 0 0 0 40 0 0 0 0 0 0

0 0 0

0 0 0

0 0 0

0 0 0

0 0 0

0 0 0

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 5 5 0 0 0 0 10 10 10 5 0 0 10 0 20 5 10 0 10 20 0 5 0 0 5 5 5 0 5 5 0 10 0 5 0 10 0 0 0 5 10 40 0 0 0 0 0 0 0 0 0 0 32 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 48 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 32 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 24 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 24 0

(7)

16 0

0 0 0 0 0 0 0 0 , 0 0 0 0 0 0 0 48

(8)

1298

0 0 0 0 0 0 0 0 0 0 0 0 15 10 10 0 0 0 0 0 0 0 0 0 VCMz = 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

Gao

0 0 0 0 5 0 0 0 0 0 0 0 0 0 0 0

15 10 10 5 40 5 15 0 15 5 0 0 0 0 0 0

0 5 0 0 5 0 10 15 0 5 0 0 0 0 0 0

0 0 0 0 15 10 0 20 15 10 5 0 0 0 0 0

0 0 0 0 0 15 20 0 0 5 10 15 5 0 0 0

0 0 0 0 15 0 15 0 0 0 0 5 15 5 5 0

0 0 0 0 5 5 10 5 0 0 5 10 0 5 5 0

0 0 0 0 0 0 5 10 0 5 0 0 0 5 0 15

0 0 0 0 0 0 0 15 5 10 0 0 0 5 10 15

0 0 0 0 0 0 0 5 15 0 0 0 0 0 0 20

0 0 0 0 0 0 0 0 5 5 5 5 0 0 0 10

0 0 0 0 0 0 0 0 5 5 0 10 0 0 10 0

0 0 0 0 0 0 0 0 0 0 15 15 20 10 0 0

(9)

In matrix (7) (VCMx ), for example, 20 in row 1 and column 2 indicates that there are 20 voxel couples of amplitude 1 neighbored by amplitude 2 along the x direction ( = 0, = 0) within the texel cube. These three VCMs are then normalized by dividing each entry with Rx = 2 8 5 9 = 720, R y = 2 9 4 9 = 648, and Rz = 2 9 5 8 = 720, respectively (see Figure 2). Finally, the algorithm reduces the normalized VCM to textural attributes using equations (4), (5), and (6). Table 1 shows the texture expressions of the example texel and indicates that VCM textures are quite sensitive to the direction in which the VCM is evaluated. Therefore, by calculating the VCM texture attributes at a voxel location, local features are extracted; spatial feature variations, on the other hand, are evaluated by sequentially and repeatedly executing the same process from voxel to voxel throughout the volume. As a result of such a running-texel processing (Figure 3), the original amplitude volume is transformed into a plurality of texture attribute volumes. These texture volumes are then visualized and interpreted individually, or they are selected and combined to produce a feature class volume using a multivariate classication algorithm (e.g., Richards, 1993; M. Taner, 1998, personal communication; Gao et al., 1998; Gao, 1999b, 2001a). To facilitate volume texture analysis in an interactive manner, I developed the VCM texture algorithms and interfaced them with a 3D seismic visualization system. Example results produced from these algorithms are shown and discussed in Figures 49. Table 1. Textural expressions of homogeneity, contrast, and randomness for the texel shown in Figure 2. Each textural attribute is evaluated along the x (intertrace in inline) direction, y (intertrace in crossline) direction, and z (intratrace in time or depth) direction. All the textural attributes are normalized to range from 0 to 1. Orientation Texture Homogeneity Contrast Randomness x 0.0353 0.0078 0.1597 y 0.0815 0.0000 0.1141 z 0.0188 0.0443 0.1791

FIG. 4. A salt canopy detected from a seed in a texture homogeneity cube. Since texture homogeneity of salt is signicantly higher than that in the surrounding areas, the whole salt body can be detected, isolated, and mapped effectively by propagating the seed from within the salt. However, it is generally difcult and time-consuming to dene the 3D geometry of the salt body directly from the amplitude volume. Because amplitude samples within the salt body are similar to and connected with those in the surrounding areas, a seed-based propagation may cause bleeding across the salt boundary and thus is not effective for automatic salt detection. Similar problem exists with mapping and isolating many other geological features using amplitude data alone. Mapping and isolating these geological features are fundamental for constructing an accurate subsurface geological model and for exploring hydrocarbons in the subsurface.

Volume Seismic Texture Analysis GEOLOGICAL IMPLICATIONS

1299

In contrast to eld geologists who map surface geology based on direct observations of outcrop structural and stratigraphic patterns, petroleum exploration geologists map subsurface geology primarily based on reection seismic patterns, which is particularly the case in frontier sedimentary basins where little direct observational data are available. Therefore, successful exploration of subsurface geology requires effective seismic pattern recognition and visualization technologies. VCM texture analysis represents one such technology that allows exploration geologists to visualize, detect, and map major geological features from a new perspective. Case examples (Figures 49) indicate that the VCM methodology signicantly enhances interpreters ability to visualize and detect major structural and stratigraphic features that may otherwise not be easily recognizable and detectable. The VCM seismic textures are indicative of several major seismic facies that are formed in diverse depositional settings. For example, in an offshore depositional setting, a domeshaped, low-amplitude seismic feature with an amplitude high at the top is typically indicative of a salt body. Such a feature has an abnormally high homogeneity (Figure 4). In a deepwater, low-energy depositional setting, a high-amplitude, laterally extensive, and coherent pattern is generally associated with sheetlike deposits of high impedance contrast. It has a relatively low homogeneity (Figure 5a), a high contrast (Figure 5b), and a high randomness (Figure 5c). A low-amplitude and laterally extensive interval typically represents a thick sequence of shale with low impedance contrast in the interval. It has a high homogeneity (Figure 5a), a low contrast (Figure 5b), and a low randomness (Figure 5c). In a turbidite system, a linear or sin-

uous feature on a map view with a concave or lenticular shape on a sectional view (Figures 6a and 6b) is generally associated with a channel. It has variable textural features, depending on the morphology, thickness, and lithology of channels, and the

FIG. 6. (a) An original amplitude section. (b) A texture homogeneity section. (c) A homogeneity cube with opacity lter applied. From homogeneity data, interpreters can effectively isolate the high-homogeneity feature (red) along a channel system by rendering transparent the low-homogeneity features (blue). It is very difcult to visualize and isolate the same features from the original amplitude volume due to the limitation of the amplitude in discriminating channels from other geological features.

FIG. 5. Three different texture attributes overlaid with the amplitude on the same section demonstrating how these attributes help distinguish and isolate intervals of different amplitude patterns. Notice, for example, that low homogeneity (blue), high contrast (red), and low randomness (blue) correspond to the laterally extensive, high-amplitude pattern A; whereas high homogeneity (red), low contrast (blue), and high randomness (red) are associated with the acoustically different pattern B (see Figure 1 for location). (a) Homogeneity (color) evaluated in the trace direction and co-rendered with original amplitude (gray). (b) Contrast (color) evaluated in the trace direction and co-rendered with original amplitude (gray). (c) Randomness (color) evaluated in the trace direction and co-rendered with original amplitude (gray).

1300

Gao

lateral/vertical partitioning patterns of the channel deposits. With massive deposits, for example, it may have a high homogeneity and a low contrast (Figures 6 and 7); with heterogeneous lithology and complex depositional geometry, it may have a low homogeneity and a high contrast (Figure 9c). A laterally more extensive and coherent feature that systematically distributes on both sides of channels or in the distal portion of a channel-fan system generally suggests levee/overbank deposits or lobes, which have different textural features than the channel-ll deposits (Figures 6 and 7). Because they are sensitive to dip and azimuth of seismic reections (e.g., Table 1; Figures 8 and 9), VCM textures are able to differentiate between seismic features having different geometry and orientations (Figures 8 and 9). Such sensitivity helps highlight and detect deformational features such as rollover anticlines or monoclines on the hanging walls of listric normal faults (Figure 8) or slumps in mass transport complexes. Similarly, such sensitivity also helps identify and map unconformities, onlaps, downlaps, and other oblique, progradational depositional features that have characteristic reection geometry in the offshore depositional setting from the shelf margin down to the basin oor. By evaluating VCM textures in a specic direction, the algorithm helps enhance faults or fractures with a preferred orientation (Figures 9a and 9b). In addition,

the direction sensitivity helps enhance the visibility of both high-angle normal or wrench faults and low-angle detachment or listric faults. This enhancement is achieved not only by the texture attribute anomalies along the faults (Figure 9), but also by textural differences across the faults (Figure 8). Such textural differences are particularly obvious across listric faults where the hanging walls have different dip and azimuth from the foot walls due to the rotational deformation that occurs in the vicinity of listric faults (Figure 8). Unlike the coherence algorithm that highlights external geometry and boundaries of geological features such as faults and channels, VCM texture analysis emphasizes their internal textures that provide hints on the facies variations within fault blocks or channel systems (Figure 9c). These internal facies variations may not be visible to the coherence algorithm. In addition, anomalous textural features (e.g., high contrast) along a fault zone (Figure 9b) enable interpreters to map and detect faults and their spatial connectivity more efciently than visual inspection and manual picking. Thus, VCM textures help dene fault zone geometry, kinematics, and relationships to the depositional facies (Figure 9), which are all important to the understanding of depositional and deformational history of a sedimentary basin.
DISCUSSION

FIG. 7. A comparison between average absolute amplitude (a) and homogeneity (b) in a horizon slice at the same stratigraphic level. To avoid a biased comparison, the same processing parameters (texel size and dimension) and a normalized color mapping function are used. Notice that the channel/levee deposits can be recognized, mapped, and detected more effectively from the homogeneity volume than from the amplitude volume.

Three-dimensional texel-based VCM texture extraction has many advantages over conventional 2D texel-based GLCM texture extraction. First, a 3D texel includes textural information from both inline and crossline directions, and allows evaluating textural features along different directions in 3D space. Thus, the 3D texel-based processing signicantly reduces interpretational biases and overcomes limitations of 2D texture processing and visual inspection. Second, a reliable extraction of the VCM seismic textures requires a sufcient number of samples that, in the 2D image space, can only be accommodated by increasing the size of the texel, thereby decreasing the resolution of the results. Whereas in the 3D image space, the accommodation problem is solved by an additional, third dimension of the texel cube, thereby signicantly enhancing the spatial resolution of the results. The structural and stratigraphic implications of VCM textures are attributable to the fact that different deformational and depositional features have characteristic internal amplitude patterns in response to the differences in acoustic impedance conguration and distribution patterns. Such internal amplitude patterns can be better dened on a volume texture basis from a 3D perspective. However, due to the complexity and nonuniqueness of seismic response to the subsurface geology, there is no simple, universal correlation between seismic textures and geological features that can be applied to any data sets in any geological settings. For example, lithology, thickness, and facies architecture of channel-levee systems in submarine turbidite systems may be distinctive in different sedimentary basins or at different times as a sedimentary basin evolves. In addition to geological complexities, variable acquisition and processing parameters, data quality, and frequency attenuation with depth may also affect textural signatures of geological features. Since prediction and classication of the subsurface geology rely on the input textural attributes, selecting textural attributes is a critical step from seismically extracted textures to a geologically meaningful prediction and

Volume Seismic Texture Analysis

1301

classication. Thus, a good understanding of geological implications of each textural attribute is fundamental to a robust interpretation and meaningful classication of the subsurface geology from seismic textures in a specic geological setting. Like any other seismic-attribute extraction algorithms that involve multiple wiggle traces in the analysis window, VCM textures are sensitive to the dip and azimuth of reection events (Table 1). Based on the comparison, I found that texture contrast is more sensitive to the direction than homogeneity and randomness. Although this sensitivity is favorable in certain aspects of structural and stratigraphic interpretation, it may also be unfavorable in interpreting depositional facies if the dip and azimuth variations are the result of postdepositional tectonic deformation. The effect can be minimized by reducing the texel size along the inline and/or crossline directions, or more effectively by searching the instantaneous dip of reection events similar to evaluating the coherence in the presence of structural dip (Marfurt et al., 1999). The VCM texture extraction methodology has a major limitation in computational efciency for high-resolution 3D seismic data. For example, for an 8-bit ( N g = 256) amplitude volume, the algorithm has to manipulate a 256 256 matrix at each sample location throughout the volume, and thus the process is computationally intensive for a large data volume that contains billions of voxels. In an attempt to solve this problem, the algorithm typically requantizes all the texels to 4-bit ( N g = 16), thereby signicantly improving the computational efciency (Haralick et al., 1973; Reed and Hussong, 1989; Gao et al.,

1998). Unfortunately, the enhancement in computational efciency is achieved at the expense of sacricing the bit resolution of the original data set. A practical solution to that problem is to run the algorithm within the interval and area of interest or on an interpreted horizon.
CONCLUSIONS

VCM seismic texture analysis, a new methodology extended from classical 2D image analysis to 3D seismic interpretation, helps visualize and detect seismic features from a different perspective than conventional seismic-attribute analysis. Such a perspective sheds new lights on certain geological features that may not be easily recognizable and detectable from the amplitude and other conventional seismic attributes. Case examples indicate that the VCM textural attributes have important implications for visualizing and mapping structural and stratigraphic features. For example, a salt body can be efciently isolated due to its high homogeneity and low contrast; a sand-lled channel can be discriminated from levee/overbank deposits based on their distinctive homogeneity and contrast. In addition, VCM textures help identify and map rollover structures or slumps produced by rotational deformation in the vicinity of listric or detachment faults. They also help identify and highlight faults with a preferred orientation and a complex geometry from a 3D perspective. Thus, VCM texture analysis signicantly enhances exploration geologists ability to visualize, isolate, and map critical seismic features that are fundamental

FIG. 8. Three different texture attributes overlaid with the amplitude on the same section demonstrating how these attributes help enhance the listric normal fault and rollover structures. Notice the differences in textures between the hanging wall (C) and the footwall (D) (see Figure 1 for location), and the distinctive textures of the rollover monocline. In this specic example, mapping and delineating both the listric fault and the rollover monocline are important for understanding migration pathways, reservoir continuity, and trapping geometry of the hydrocarbon system. (a) Homogeneity (color) evaluated in the crossline direction and co-rendered with original amplitude (gray). (b) Contrast (color) evaluated in the crossline direction and co-rendered with original amplitude (gray). (c) Randomness (color) evaluated in the crossline direction and co-rendered with original amplitude (gray).

1302

Gao

to robust geological interpretation and successful hydrocarbon exploration.


ACKNOWLEDGMENTS

I started this study at Exxon Production Research Company (in 1997) and developed it at Marathon Oil Corporation (in 1998). I am grateful to Marathon management for permission to publish this work. Thanks are due to Sharon Crawford, Tom Evans, and Steve Peterson for their support and suggestions in this study. I used the application program interface (API) functions from Magic Earth Inc. and Paradigm Geophysical Inc. in the development of the VCM texture extraction and visualization algorithms. The 3D seismic data sets used in this publication are provided courtesy of the Bureau of Economic Geology, Austin, Texas, and Seitel, Houston, Texas. Journal reviews by the associate editor Kurt J. Marfurt and two anonymous reviewers helped improve the quality of the paper.
REFERENCES Bahorich, M., and Farmer, S., 1995, 3-D seismic discontinuity for faults and stratigraphic features: The coherence cube: The Leading Edge, 14, 10531058. Gao, D., 1999a, The rst-order and the second-order seismic textures: AAPG Abstracts with Programs, 8, A45. 1999b, 3-D VCM seismic textures: A new technology to quantify seismic interpretation: 69th Ann. Internat. Mtg., Soc. Expl. Geophys., Expanded Abstracts, 10371039. 2001a, Method for analyzing and classifying three dimensional seismic information: US Patent 6 226 596. 2001b, Seismic texture visualization and interpretation: An overview, in Graul, M., and Taner, M. T., Eds., Reservoir resolution through comprehensive use of seismic data attributes: Soc. Expl. Geophys. and Geophysical Society of Houston Spring Symposium, paper 2.5. 2002, Seismic textures aid exploration: Offshore, 62, No. 9, 65. Gao, D., Hurst, S. D., Karson, J. A., Delaney, J., and Spiess, F., 1998, Computer-aided interpretation of side-looking sonar images from the eastern intersection of the Mid-Atlantic Ridge with the Kane transform: J. Geophys. Res., 103, 2099721014. Haralick, R. M., Shanmugam, K., and Dinstein, I., 1973, Textural features for image classication: IEEE Tran. Systems, Man, and Cybernetics, SMC-3, 610621. Marfurt, K. J., Sudhaker, V., Gersztenkorn, A., Crawford, K. D., and Nissen, S. E., 1999, Coherency calculations in the presence of structural dip: Geophysics, 64, 104111. Reed, T. B., and Hussong, D., 1989, Digital image processing techniques for enhancement and classication of SeaMARC II side-scan sonar imagery: J. Geophys. Res., 94, 74697490. Richards, J. A., 1993, Remote sensing digital image analysis. SpringerVerlag. Taner, M. T., and Sheriff, R. E., 1977, Application of amplitude, frequency, and other attributes to stratigraphic and hydrocarbon determination, in Payton, C. E., Ed., Seismic stratigraphyapplications to hydrocarbon exploration: AAPG Memoir, 26, 301327. Taner, M. T., Schuelke, J. S., ODoherty, R., and Baysal, E., 1994, Seismic attributes revisited: 64th Ann. Internat. Mtg., Soc. Expl. Geophys., Expanded Abstracts, 11041106. Vinther, R., Mosegaard, K., Abatzis, I., Andersen, C., Vejbaek, O. V., If, F., and Nielsen, P. H., 1996, 3D seismic texture classication: SPE/Norwegian Petr. Soc. Europe. 3-D Reservoir Modeling Conf. Proc., 5763. Weszka, J. S., Dyer, C. R., and Rosenfeld, A., 1976, A comparative study of texture measures for terrain classication: IEEE Trans. Systems, Man, and Cybernetics, SMC-6, 269285. Zhang, Z., and Simaan, M., 1989, Knowledge-based reasoning in seisis: A rules-based system for interpretation of seismic sections based on texture, in Aminzadeh, F., and Simaan, M., Eds., Expert systems in exploration: Soc. Expl. Geophys., 141159.

FIG. 9. (a) Texture contrast along the x direction (east-west). (b) Texture contrast along the y direction (north-south). (c) Texture contrast along the z direction (vertical). (d) Interpretation. Notice that contrast evaluated along the x direction (a) helps highlight the north-south trending fractures f1, whereas contrast evaluated along the y direction (b) highlights the primary east-west trending fault and fractures f2, and contrast evaluated along the z direction (c) helps identify depositional features such as channels. The geometric relationship between the major fault (F) and the two conjugate fractures (f1 and f2) suggests left-lateral displacement along the fault. Such an interpretation is consistent with the offset of the prefault depositional facies across the fault. Also note that there are at least two stages of channel development (c). The channels to the east was developed prior to the fault displacement and were subsequently truncated and offset left laterally by the fault. The channels to the west were developed after the major fault displacement and ran across the fault. These interpretations are shown in (d) based on the observations from the texture data shown in (a), (b), (c), and the regional geology of the study area.

You might also like