Open Access

New parameter of roundness R: circularity corrected by aspect ratio

Progress in Earth and Planetary Science20163:2

DOI: 10.1186/s40645-015-0078-x

Received: 14 January 2015

Accepted: 27 December 2015

Published: 20 January 2016

Abstract

In this paper, we propose a new roundness parameter R, to denote circularity corrected by aspect ratio. The basic concept of this new roundness parameter is given by the following equation:

$$ R=\mathrm{Circularity} + \left({\mathrm{Circularity}}_{\mathrm{perfect}\ \mathrm{circle}}-{\mathrm{Circularity}}_{\mathrm{aspect}\ \mathrm{ratio}}\right) $$

where Circularityperfect circle is the maximum value of circularity and Circularityaspect ratio is the circularity when only the aspect ratio varies from that of a perfect circle. Based on tests of digital circle and ellipse images using ImageJ software, the effective sizes and aspect ratios of such images for the calculation of R were found to range between 100 and 1024 pixels, and 10:1 to 10:10, respectively. R is thus given by

$$ R = {\mathrm{C}}_{\mathrm{I}} + \left(0.913-{\mathrm{C}}_{\mathrm{AR}}\right) $$

where CI is the circularity measured using ImageJ software and CAR is the sixth-degree function of the aspect ratio measured using the same software. The correlation coefficient between the new parameter R and Krumbein’s roundness is 0.937 (adjusted coefficient of determination = 0.874). Results from the application of R to modern beach and slope deposits showed that R is able to quantitatively separate both types of material in terms of roundness. Therefore, we believe that the new roundness parameter R will be useful for performing precise statistical analyses of the roundness of particles in the future.

Keywords

Aspect ratio Circularity ImageJ software Krumbein’s visual roundness Roundness

Background

Many studies have investigated particle shape in the natural world, mostly based on the definitions of sphericity and roundness of rock particles proposed by Wadell (1932). Previous studies into particle shape have been discussed in a series of review articles (e.g., Barrett 1980; Clark 1981; Winkelmolen 1982; Diepenbroek et al. 1992; Blott and Pye 2008), and in general, such studies have mainly taken one of two approaches to understanding particle shape. The first is a simple method that involves the examination of visual images of particle grains (e.g., Krumbein 1941; Rittenhouse 1943; Powers 1953; Pettijohn 1957; Lees 1963). Determining roundness using the visual roundness chart proposed by Krumbein, which is further extended in this paper, is one of the most widely employed methods. However, such a method merely compares visual images, and therefore, the derived roundness values are not strictly quantitative. The second approach involves the quantitative determination of various shape parameters, and many evaluation methods have been designed to obtain relevant shape parameters (e.g., Schwarcz and Shane 1969; Orford and Whalley 1983; Diepenbroek et al. 1992; Yoshimura and Ogawa 1993; Vallejo and Zhou 1995; Bowman et al. 2001; Itabashi et al. 2004; Drevin 2007; Blott and Pye 2008; Lira and Pina 2009; Roussillon et al. 2009; Arasan et al. 2011; Suzuki et al. 2013, 2015). Both approaches, however, involve the analysis of each individual particle, and therefore, production of several thousand to several tens of thousands of shape parameters for reliable analysis is time consuming. Therefore, owing to the extensive time requirements and effort required for both approaches, neither is widely used, and there remains a need for an easy statistical method to derive parameters of particle shape.

In this study, we define a new roundness parameter, R, to denote the circularity corrected by aspect ratio, and present a case study of R calculation using ImageJ software (ver. 1.47q) released from the US National Institute of Health (Abramoff et al. 2004; Schneider et al. 2012). This represents a fairly simple method that helps to overcome the shortcomings of the previously published methods discussed above.

Basic concept of this study

The proposed concept for the new roundness parameter is quite simple: It is a correction of circularity using the aspect ratio of particles. The definition of circularity, corresponding to that of parameter K defined by Cox (1927), is given as follows:
$$ \mathrm{Circularity} = 4\pi \cdot \frac{\mathrm{Area}}{{\mathrm{Perimeter}}^2} $$
(1)
This indicates that circularity can be altered in two ways: by changes in area and by changes in the perimeter of a particle. To consider this, an ideal perfect circle (true circle) is assumed. If the area and perimeter do not change, then circularity is constant. However, if only the perimeter increases and the area does not change (Fig. 1a, towards the right), then circularity decreases. An increase in the perimeter length therefore represents a decrease in the roundness of the particle. It should be noted that in this paper, we use the term “roundness” to refer to the presence or absence of surface irregularities. Therefore, with a decrease in roundness, circularity also decreases. In comparison, in the case of the ellipses skewed from a circle, if only the area decreases and the perimeter does not change, then circularity can also be seen to decrease (Fig. 1b). A decrease in area in this way represents an increase in the aspect ratio of the particle image. Therefore, with a increase in the aspect ratio, the circularity also decreases. In Fig. 1b, however, the transformed images in the center and on the right still show a high degree of roundness. Consequently, it should be possible to determine roundness if circularity can be corrected using aspect ratio. In other words, if it is possible to combine the difference in circularity value due to aspect ratio with the circularity itself, then the aspect ratio-corrected circularity can be used to represent roundness. Thus, our roundness parameter R can be defined by the following equation:
Fig. 1

Basic concept of transformation from a perfect circle. Narrow solid lines denote perfect circles before transformation. a Only the perimeter increases; the area does not change. b Only the area decreases; the perimeter does not change

$$ R=\mathrm{Circularity} + \left({\mathrm{Circularity}}_{\mathrm{perfect}\ \mathrm{circle}}-{\mathrm{Circularity}}_{\mathrm{aspect}\ \mathrm{ratio}}\right) $$
(2)

where Circularity is the value defined by Eq. (1), Circularityperfect circle is the maximum value of circularity, and Circularityaspect ratio is the circularity when only the aspect ratio varies from that of a perfect circle.

Case study of parameter R calculation using ImageJ software

In this chapter, we present a specific case study, demonstrating the calculation of the parameter R defined in the previous chapter, using ImageJ software.

Methods

Test digital images were produced using Adobe Photoshop CS4 and Adobe Illustrator CS4. Shape parameters, including area, perimeter, circularity, aspect ratio, major axis length, and minor axis length, were measured from the test digital images using ImageJ software (ver. 1.47q). To validate the effectiveness of R defined in this paper, digital images of Krumbein’s original visual images (Krumbein 1941) were captured using a Fuji Xerox ApeosPort-IV C7780 scanner with a resolution of 600 × 600 dots per inch and a grayscale color profile; they were saved in a TIF image file format.

Grain-size distributions of modern slope and beach sediments were measured using a sieve ranging from −5.0 to 4.0 phi, with 0.5 phi intervals. The first quartile (twenty-fifth percentile), second quartile (median), and third quartile (seventy-fifth percentile) of grain size were obtained for each from the cumulative curves.

To obtain the new roundness parameter R value for these modern slope and beach sediments, we used an Olympus TG-1 digital camera to acquire digital images of particles in each grain-size class. Image analysis was then conducted separately for each grain-size fraction. The measurement of grain sizes in this analysis ranged from values equal to or coarser than 1.0 phi, with 0.5 phi intervals, with the finer limit (1.0 phi) of the measurement range determined by the limitations of the Olympus TG-1. This range was sufficient for comparing the sediments in this study, because of the coarseness of the material. For imaging, the particles were laid out on a transparent board. As the minor c-axes of the particles in this layout were nearly perpendicular to the board, we assumed that an imaginary plane perpendicular to the c-axis, which included the major and intermediate a- and b-axes, respectively, was parallel to the board. To obtain sharp silhouettes of particles, the light source was placed on the opposite side of the digital camera, allowing intentional capture of backlit images. The major lengths of the silhouettes were then adjusted to be more than 100 pixels. The digital images taken by Olympus TG-1 were transferred to the ImageJ software and processed into binary images. The circularity and aspect ratio of the silhouettes in the digital binary images were then measured using the ImageJ software, and the R values were obtained using Eq. (8) described below. The obtained R values for each grain-size class were integrated into a total R distribution for the individual samples using the weight percent of each grain size class. The calculated R distributions thus ranged from 0.400 to 0.925 with 0.025 intervals. The first quartile, median, and third quartile of R were calculated using Microsoft Excel 2007 software.

Validation of the effective resolution of digital images using ImageJ software

A digital image is an aggregate of pixels, which are the minimum units of the image. Hence, there should be an error in shape parameter values between the geometrically obtained true values and those calculated from the digital image. Therefore, to obtain the most effective size of digital images for shape analysis, we examined the errors in basic shape parameters, including area, perimeter, and aspect ratio, using ImageJ software.

Area and perimeter

First, the area (A I), perimeter (P I), and major and minor axis lengths of the fit ellipses (2·r; both lengths are equal) of twelve test circle images with diameter lengths of 1 to 21 pixels, produced using Adobe Photoshop CS4, were calculated using ImageJ software (Fig. 2). To calculate the area and perimeter, the measuring algorithms of the ImageJ software were sought from the manual. However, there were no detailed descriptions of the algorithms, so we investigated the value determinations ourselves. The obtained algorithms are therefore as follows. A I is equal to the total number of pixels in a grain. For test circle images with a diameter length of 1 to 2 pixels, P I was calculated by determining the geometric mean of the numbers of pixels in the images and the numbers of circumscribed pixels in the images. In contrast, for test circle images with diameter lengths greater than 2 pixels, P I was the sum of the marginal pixels, in which the sizes of pixels located at the corners are assumed to be 20.5, while those of other pixels is 1. The major axis length of the test circle images was 2·r, because all of the test images comprised perfect circles. The error (ΔA) between the area measured using ImageJ software (A I) and the area calculated from radius length (A), and the error (ΔP) between P I and the perimeter obtained by the geometric procedure (P), were therefore determined as follows (Table 1):
Fig. 2

Test circle images with diameters of 1 to 1448 pixels

Table 1

Results of error of area and perimeter for various sizes of test circle images

Width of test circle image (pixel)

A I

P I

r

A

P

ΔA (%)

ΔP (%)

Area measured by ImageJ (pixel)

Perimeter measured by ImageJ (pixel)

Major and minor axis length measured by ImageJ

Area calculated by p·r 2

Perimeter calculated by 2·p·r

1

1

2.828

1.128

0.999328

3.543716

0.067260

−20.196760

2

4

5.657

2.257

4.000856

7.090573

−0.021393

−20.218015

3

9

9.657

3.385

8.999267

10.634289

0.008145

−9.189979

4

12

11.314

3.909

12.001102

12.280483

−0.009184

−7.870074

6

32

19.314

6.383

31.999226

20.052782

0.002417

−3.684186

8

52

24.971

8.137

52.001806

25.563134

−0.003472

−2.316360

11

97

34.627

11.113

96.995686

34.912512

0.004447

−0.817792

16

208

52.284

16.274

208.007222

51.126268

−0.003472

2.264456

23

421

75.598

23.152

420.985191

72.734138

0.003518

3.937439

32

804

104.569

31.995

803.996244

100.515236

0.000467

4.032985

45

1581

147.196

44.866

1580.973153

140.950667

0.001698

4.430865

64

3196

211.48

63.791

3196.013548

200.405295

−0.000424

5.526154

90

6320

296.735

89.704

6319.946608

281.813369

0.000845

5.294863

128

12,796

422.96

127.642

12796.081336

400.999086

−0.000636

5.476550

181

25,565

597.47

180.417

25564.935720

566.796604

0.000251

5.411711

254

51,104

842.607

255.084

51104.156942

801.369854

−0.000307

5.145832

360

102,252

1193.97

360.82

102251.807881

1133.549225

0.000188

5.330229

510

204,520

1690.871

510.297

204520.017549

1603.144973

−0.000009

5.472121

722

408,932

2391.253

721.574

408932.420696

2266.891106

−0.000103

5.486011

1020

818,196

3381.743

1020.666

818195.500707

3206.516140

0.000061

5.464712

1442

1,636,024

4893.536

1443.278

1636024.391899

4534.190619

−0.000024

7.925238

p = 3.141592

$$ \varDelta A=\left(\frac{A_{\mathrm{I}}-A}{A}\right)\cdotp 100 $$
(3)
$$ \varDelta P = \left(\frac{P_{\mathrm{I}}-P}{P}\right)\cdotp 100 $$
(4)
From our calculations, the absolute error for the area (|ΔA|) was found to decrease with an increase in the diameter of the test circle images (Table 1 and Fig. 3). All 21 |ΔA| values obtained in this test were below 0.1 %, which is sufficiently small to assume accuracy. These values are therefore sufficiently reliable for use in shape analysis. In contrast, the absolute perimeter error |ΔP| was approximately 20 % for diameters of 1 to 2 pixels, but decreased with an increase in the diameter of the test circle images. |ΔP| attained a minimum of approximately 0.8 % at a diameter of 11 pixels; however, the error increased again to approximately 8 % for diameters of 16 to 1442 pixels. For diameters of 64 to 1024 pixels, the |ΔP| value remained constant at around 5 % (Table 1 and Fig. 3).
Fig. 3

Plot of the diameter of a digital circle image (pixels) and error (%) in the area and perimeter estimations

Aspect ratio

In this section, we examine the relationship between the aspect ratio measured using ImageJ software (ARI) and the widths of test ellipse images (Fig. 4), which were obtained by subjecting the test circle images in Fig. 2 to 10 % aspect ratio deformation intervals in Adobe Photoshop CS4. The ImageJ software defines the aspect ratio as follows:
Fig. 4

Test circle/ellipse images with diameters/widths of 1 to 1448 pixels and aspect ratios of 10/10 to 10/1

$$ {\mathrm{AR}}_{\mathrm{I}} = \frac{\mathrm{major}\ \mathrm{axis}\ \mathrm{length}\ \mathrm{of}\ \mathrm{approximate}\ \mathrm{ellipse}}{\mathrm{minor}\ \mathrm{axis}\ \mathrm{length}\ \mathrm{of}\ \mathrm{approximate}\ \mathrm{ellipse}} $$
(5)
Therefore, the aspect ratio is equal to one for a perfect circle and increases with an increase in deformation. To validate this relationship for different image sizes, we prepared test circle images with diameters of 20–10.5 (1 to 1448) pixels, and test ellipse images were obtained from the test circle images through 10 % deformation intervals in height (Fig. 4). As such, AR I showed a range of values, controlled by the widths of the test circle/ellipse images (Fig. 5). Notably, the test circle/ellipse images with diameters or widths of less than 100 pixels are unstable, as shown in Fig. 5. However, the widths of the test circle/ellipse images greater than or equal to 100 pixels can be seen to remain constant.
Fig. 5

Plot of the diameter of a digital circle image (pixels) and the aspect ratio calculated using ImageJ (ARI)

Effective size range of digital images in this study

The above results can be summarized as follows: (1) The |ΔA| values are sufficiently small to assume accuracy for all test circle images with diameters ranging from 1 to 1442 pixels; (2) The |ΔP| value is constant, at approximately 5 %, for test circle images with diameters of 64 to 1024 pixels; and (3) The ARI value for the test circle/ellipse images with widths greater than or equal to 100 pixels, and a true aspect ratio greater than or equal to 10:1, remains constant. Consequently, in this paper, we consider the effective size and aspect ratios of digital images for shape analysis using ImageJ software, to be 100 to 1024 pixels and 10:1 to 10:10, respectively.

Relationship between aspect ratio and circularity in ImageJ software

Circularity in ImageJ software

Circularity can be calculated as a shape parameter index in the ImageJ software. The definition of circularity (C I) in the ImageJ software is as follows:
$$ {C}_{\mathrm{I}} = 4\pi \cdot \frac{A_{\mathrm{I}}}{{P_{\mathrm{I}}}^2} $$
(6)

where A I and P I are the area and perimeter measured using ImageJ (ImageJ User 2012), respectively. This therefore implies that C I is directly determined by A I and P I. For instance, if their two different P I values are provided for digital images with the same A I values, the image showing high circularity will have a shorter perimeter than that of the other image.

Relationship between circularity and aspect ratio in ImageJ software

When considering roundness as a shape parameter, the degree of roundness of a deformed circle (ellipse) image should be the same as that of a perfect circle image. Different values of C I will therefore correspond to changes in aspect ratio, and C I alone should not be used as a roundness shape parameter. For this reason, the relationship between C I and ARI is examined in this section, and we attempt to correct C I using ARI.

First, a test circle image with a diameter of 100 pixels was produced using Adobe Illustrator CS4. Then, 91 test ellipse images were obtained by deforming a test circle image, in 0.1 % height intervals (Fig. 6). From the shape analysis of circularity and aspect ratio using ImageJ software, the triadic relationship between the true aspect ratio, C I, and ARI could be obtained (Table 2). ARI versus C I is plotted in Fig. 7. It should be noted that the maximum C I value is 0.913 (Table 2), as ImageJ is unable to output 1.0 as a maximum value for these test images, because the digital images comprise an aggregate of pixels and include errors. We used a polynomial regression to analyze the relationships. From this analysis, a sixth-degree polynomial was obtained for the relationship, with a strong correlation (r = 0.999836119, p < 0.005, adjusted coefficient of determination = 0.999648855; solid curve in Fig. 7). Thus, the estimated regression equation for the survey line is
Fig. 6

Test circle/ellipse images with aspect ratios of 10/10 to 10/1 and diameters/widths of 100 pixels

Table 2

CI and ARI values calculated using ImageJ applied to the test images of Fig. 6

aspect ratio of test ellipse image

CI

ARI

aspect ratio of test ellipse image

CI

ARI

aspect ratio of test ellipse image

CI

ARI

10:10.0

0.911

1.000

10:6.9

0.871

1.449

10:3.8

0.668

2.626

10:9.9

0.905

1.010

10:6.8

0.867

1.471

10:3.7

0.654

2.711

10:9.8

0.912

1.020

10:6.7

0.860

1.493

10:3.6

0.646

2.775

10:9.7

0.913

1.031

10:6.6

0.856

1.515

10:3.5

0.634

2.860

10:9.6

0.912

1.042

10:6.5

0.855

1.539

10:3.4

0.621

2.936

10:9.5

0.913

1.054

10:6.4

0.851

1.560

10:3.3

0.609

3.038

10:9.4

0.909

1.064

10:6.3

0.845

1.588

10:3.2

0.599

3.116

10:9.3

0.910

1.075

10:6.2

0.842

1.612

10:3.1

0.583

3.230

10:9.2

0.909

1.087

10:6.1

0.836

1.641

10:3.0

0.570

3.335

10:9.1

0.910

1.099

10:6.0

0.832

1.665

10:2.9

0.558

3.439

10:9.0

0.904

1.110

10:5.9

0.826

1.694

10:2.8

0.545

3.573

10:8.9

0.906

1.123

10:5.8

0.822

1.724

10:2.7

0.530

3.696

10:8.8

0.904

1.136

10:5.7

0.815

1.756

10:2.6

0.516

3.847

10:8.7

0.909

1.149

10:5.6

0.810

1.785

10:2.5

0.503

3.984

10:8.6

0.902

1.161

10:5.5

0.807

1.816

10:2.4

0.486

4.172

10:8.5

0.905

1.175

10:5.4

0.798

1.851

10:2.3

0.470

4.338

10:8.4

0.898

1.190

10:5.3

0.791

1.887

10:2.2

0.453

4.549

10:8.3

0.901

1.205

10:5.2

0.777

1.924

10:2.1

0.438

4.755

10:8.2

0.897

1.219

10:5.1

0.780

1.959

10:2.0

0.422

5.009

10:8.1

0.898

1.234

10:5.0

0.773

2.002

10:1.9

0.405

5.236

10:8.0

0.891

1.250

10:4.9

0.761

2.039

10:1.8

0.386

5.570

10:7.9

0.893

1.265

10:4.8

0.758

2.082

10:1.7

0.369

5.860

10:7.8

0.886

1.281

10:4.7

0.748

2.127

10:1.6

0.350

6.260

10:7.7

0.892

1.298

10:4.6

0.741

2.173

11:1.5

0.334

6.636

10:7.6

0.886

1.315

10:4.5

0.733

2.221

12:1.4

0.31

7.15

10:7.5

0.89

1.33

10:4.4

0.72

2.27

13:1.3

0.29

7.64

10:7.4

0.88

1.35

10:4.3

0.72

2.32

14:1.2

0.27

8.34

10:7.3

0.88

1.37

10:4.2

0.7

2.38

15:1.1

0.25

8.98

10:7.2

0.88

1.39

10:4.1

0.7

2.44

16:1.0

0.23

10.05

10:7.1

0.88

1.41

10:4.0

0.68

2.5

   

10:7.0

0.87

1.43

10:3.9

0.68

2.57

   
Fig. 7

Plot of ARI and C I values of the test images, calculated using ImageJ software. Solid line represents a sixth-degree polynomial

$$ {C}_{\mathrm{I}}=0.826261+0.337479\cdotp {\mathrm{AR}}_{\mathrm{I}}-0.335455\cdotp {{\mathrm{AR}}_{\mathrm{I}}}^2+0.103642\cdotp {{\mathrm{AR}}_{\mathrm{I}}}^3 - 0.0155562\cdotp {{\mathrm{AR}}_{\mathrm{I}}}^4 + 0.00114582\cdotp {{\mathrm{AR}}_{\mathrm{I}}}^5-0.0000330834\cdotp {{\mathrm{AR}}_{\mathrm{I}}}^6 $$
(7)

where C I is the circularity calculated using ImageJ software and ARI is the aspect ratio calculated using the same software. This indicates that C I is a sixth-degree polynomial of ARI when test ellipse images are made from the deformation of a perfect circle image with 0.1 % intervals in height. We therefore refer to the C I value newly derived from this regression equation as C AR.

Calculating roundness parameter R using ImageJ software

Equation (7) implies that the circularity of a perfect circle changes with varying aspect ratio. This means that if a shape’s perimeter is highly rounded, then the degree of roundness should also be close to 0.913. Therefore, after the correction of circularity by aspect ratio, which represents the addition of the difference between 0.913 and C AR to C I, it is possible to calculate the circularity without considering aspect ratio. The corrected circularity is therefore considered the new roundness parameter (R) and can be presented as follows:
$$ R={C}_{\mathrm{I}}+\left(0.913 - {C}_{\mathrm{AR}}\right) $$
(8)

When using R in particle shape analysis, the roundness values can be easily handled as numerical data. For instance, the first quartile (twenty-fifth percentile), second percentile (median), and third quartile (seventy-fifth percentile) of the roundness of a large number of particle grains can be easily and quickly examined.

Validation of R using Krumbein’s pebble images for visual roundness

In order to validate the effectiveness of R defined in this paper, the R values of Krumbein’s visual images were calculated and examined. Figure 8 plots Krumbein’s roundness (R K ) against the R values calculated using the method defined in this study. From this analysis, a second-degree polynomial was obtained for the relationship, with strong correlation (r = 0.936801380, p < 0.005; adjusted coefficient of determination = 0.874458283; Fig. 8), and the estimated regression equation for the survey line is therefore
Fig. 8

Plot of R K and R values derived in this study. Solid line represents the quadratic regression

$$ {R}_K=6.9940-20.575\cdotp R + 15.349\cdotp {R}^2 $$
(9)

This implies that R is an effective parameter for roundness.

Correlation analyses have been conducted, comparing shape parameters and the roundness of Krumbein’s visual images, by many previous authors (Table 3). These previously published shape parameters were calculated through various methods, including Fourier analysis (Mi: Itabashi et al. 2004), fractal analysis (D: Vallejo and Zhou 1995; FD: Itabashi et al. 2004), and computer-assisted geometrical analysis (FU: Itabashi et al. 2004; rW: Roussillon et al. 2009). Strong correlation coefficients were obtained between these shape parameters and the Krumbein’s (1941) roundness. In particular, Mi, FD, and rW all had high correlation coefficients of more than 0.9 (0.940, 0.939, and 0.919, respectively). Similar to these studies, the shape parameter R defined in this study also exhibits a high correlation coefficient (0.937). This demonstrates that R is a suitable parameter for discussing the roundness of particle grains as Mi, FD, and rW.

However, R has an advantage over previously defined parameters in that it can be used to easily obtain roundness values using widely available software (such as ImageJ software). Consequently, the new roundness parameter R can be expected to have a significant effect on future statistical analyses of roundness. The roundness parameter R is also advantageous as it can be applied as a part of simple new field studies into clastic grain shapes, which is not the case for other methods such as fractal dimensions or Fourier descriptors. This simple approach to calculating the circularity corrected by aspect ratio has a great potential that can advance research in a wide variety of scientific fields.

Applying R to modern deposits using ImageJ software

In this section, we apply our shape parameter to digital images of samples of sedimentary materials collected from modern beaches and slopes in the Masaki area of Eastern Japan (Figs. 9 and 10). The lithology of this region is mainly Cretaceous rhyolite, dacite, sandy siltstone, sandstone, and conglomerate (Shimazu et al. 1970). The beach deposits consist primarily of coarse-grained sands to granules with pebbles, which are well-abraded by wave action on the beach (Fig. 10a). The slope deposits in this area comprise heavily weathered Cretaceous basement rocks, which occur as angular granule- to pebble-sized clasts (Fig. 10b). Thus, the slope deposits can be considered immature clastics, while the beach deposits represent more mature clastic material. We selected these two types of materials for roundness analysis because they differ markedly in particle roundness, but both have a fairly coarse grain-size distribution.
Fig. 9

Location of the study area. a Large-scale map of study area in Northeastern Japan and facing the Pacific Ocean. b Detailed map with sampling location in the Masaki area of Iwate Prefecture, Northeastern Japan. Location map is based on “Taro,” the 1:25,000 topographic map from the Geographical Survey Institute (GSI) of Japan. Solid circle and solid square indicate beach and slope deposits, respectively

Fig. 10

Photographs and binary images of sediment particles. These were obtained from beach and slope environments and grain size ranges from −1.5 to 2.0 phi. Photographs of a beach sediments and b slope sediments (scoop length is 82 cm); photographs of prepared samples of c beach sediments and d slope deposits; binary images processed by ImageJ of e beach sediments and f slope deposits

Grain size distribution

A total of eight sedimentary samples were selected from beach and slope environments for this study, consisting of four beach deposits, referred to as B1, B2, B3, and B4, and four slope deposits, referred to as S1, S2, S3, and S4. Before shape analysis, their grain-size distributions were measured using a sieve ranging from −5.0 to 4.0 phi with 0.5 phi intervals (Fig. 11). Median grain sizes of the deposits in each environment were 0.33 (B1), −0.50 (B2), −1.25 (B3), and −0.50 (B4) phi for the beach deposits, and −3.18 (S1), −5.00 (S2), −4.77 (S3), and −2.47 (S4) phi for the slope deposits. The range between the first and third quartiles in the grain size distribution can be considered a proxy for the degree of sorting. Thus, these ranges in each environment were 0.821 to 0.869 (B1), −1.34 to 0.50 (B2), −1.93 to −0.40 (B3), and −1.70 to 0.16 (B4) phi for the beach deposits and −3.96 to −1.96 (S1), −5.00 to −4.94 (S2), −5.00 to −4.30 (S3), and −3.46 to −1.33 (S4) phi for the slope deposits. Together, these data indicate that the beach deposits were finer than the slope deposits but had similar degree of sorting.
Fig. 11

Grain-size distributions and R distributions of beach deposits and slope deposits

R distribution

The R values were measured using ImageJ software, following the above-described methodology. The measured R distributions in these two deposits are shown in Fig. 12. The median R of deposits in each environment was 0.847 (B1), 0.860 (B2), 0.865 (B3), and 0.866 (B4) for the beach deposits and 0.764 (S1), 0.778 (S2), 0.746 (S3), and 0.784 (S4) for the slope deposits. The range between the first and third quartiles of R in the deposits of each environment were 0.821 to 0.869 (B1), 0.831 to 0.860 (B2), 0.835 to 0.889 (B3), and 0.838 to 0.889 (B4) for the beach deposits and 0.703 to 0.804 (S1), 0.759 to 0.792 (S2), 0.726 to 0.776 (S3), and 0.741 to 0.815 (S4) for the slope deposits. Together, these data indicate that the beach deposits were more rounded than the slope deposits.
Fig. 12

Plot of median R versus median grain size for modern sedimentary materials. Solid circles and solid squares indicate beach and slope deposits, respectively. Vertical and horizontal error bars represent the ranges between the first and third quartiles

Table 3

Correlation and regression expression between shape parameters values and the Krumbein’s visual images

Shape parameters

Correlation coefficient

   

Regression expressions

References

Individual valuesa (n = 81)

Adjusted coefficient of determination of individual values (n = 81)

Mean values (n = 9)

Adjusted coefficient of determination of mean values (n = 9)

  

D

0.778b

0.549b

D = 1.0541 − 0.0335·R K

Vallejo and Zhou (1995)

R K  = 19.255 − 18.079·D c

Mi

0.940

Mi = 28.38 − 46.18 R K  + 21.71 R K 2

Itabashi et al. (2004)

FD

0.939

FD = 1.03655 − 0.05799·R K  + 0.02698·R K 2

Itabashi et al. (2004)

FU d

0.857

FU = 0.0736 + 0.264·R K

Itabashi et al. (2004)

rW

0.919

0.992

 

Roussillon et al. (2009)

R

0.937

0.874

0.995

0.987

R K  = 6.9940 − 20.575·R + 15.349·R 2

This study

aThe shape factor for each group is the average of nine shape factors corresponding to the nine particles forming each group in Krumbein’s pebble images

bThis parameter is recalculated values to the third decimal place by the author

cThis expression is recalculated by the author

dThis parameter is the same definition as circularity on ImageJ

Comparison between beach and slope deposits using R and grain-size distributions

Through the examination of both R and grain-size distributions (Fig. 12), distinctive differences between beach and slope deposits are revealed. In order to compare the characteristics of the different deposits, a plot of the mean values and ranges between the first and third quartiles is shown for all samples in Fig. 12. This diagram shows clearly that the areas in which the beach deposits and slope deposits plot are completely separate. The beach deposits had high R values and ranged from coarse-grained sands to granules. In contrast, the slope deposits had low R values and ranged in size from granules to pebbles. These distinct variations imply that there is a significant difference in cumulative energy between the two deposit types. The beach deposits comprise particles that are highly abraded by wave action and beach drift transport, whereas the slope deposits were comparatively unaffected by such physical abrasion. Therefore, the new roundness parameter R can be considered helpful for the study of sedimentary processes and the estimation of particle origins.

Conclusions

In this study, we propose a new particle roundness parameter, R, which can be defined as the circularity corrected by the aspect ratio, and we demonstrate the calculation of this parameter from particle images, using ImageJ software. The results of this study can be summarized as follows:
  1. 1.

    The basic concept of the new roundness parameter R can be defined as:

     
$$ R=\mathrm{Circularity} + \left({\mathrm{Circularity}}_{\mathrm{perfect}\ \mathrm{circle}}-{\mathrm{Circularity}}_{\mathrm{aspect}\ \mathrm{ratio}}\right) $$
where Circularityperfect circle is the maximum value of circularity and Circularityaspect ratio is the circularity when only the aspect ratio varies from that of a perfect circle.
  1. 2.

    The effective diameter of a digital image suitable for R calculations ranges from 100 to 1024 pixels, based on shape analysis of test circle images of various sizes using ImageJ software.

     
  2. 3.

    The effective aspect ratio of digital images for R calculations ranges from 10:1 to 10:10, based on shape analysis for various test circle and ellipse images in ImageJ software.

     
  3. 4.

    Given that a digital image is of an appropriate size, circularity (C AR) is given by a sixth-degree polynomial with respect to aspect ratio (ARI):

     
$$ {C}_{\mathrm{AR}}=0.826261+0.337479\cdotp {\mathrm{AR}}_{\mathrm{I}}-0.335455\cdotp {{\mathrm{AR}}_{\mathrm{I}}}^2+0.103642\cdotp {{\mathrm{AR}}_{\mathrm{I}}}^3 - 0.0155562\cdotp {{\mathrm{AR}}_{\mathrm{I}}}^4 + 0.00114582\cdotp {{\mathrm{AR}}_{\mathrm{I}}}^5-0.0000330834\cdotp {{\mathrm{AR}}_{\mathrm{I}}}^6 $$
  1. 5.

    The new roundness parameter R is thus defined as:

     
$$ R={C}_{\mathrm{I}}+\left(0.913-{C}_{\mathrm{AR}}\right) $$
where C I is the circularity measured using ImageJ software.
  1. 6.

    Validation of R using the pebble images for visual roundness provided by Krumbein (1941) reveals a strong correlation coefficient (r = 0.937) between Krumbein’s roundness and R.

     
  2. 7.

    Based on the application of R to modern beach and slope deposits, we can confirm that the new roundness parameter R represents a useful new tool in the analysis of particle shape.

     

Declarations

Acknowledgements

We thank the two anonymous reviewers who provided constructive comments and helpful suggestions. This research was partly supported by Grants-in-Aid for Scientific Research from the Japan Society for the Promotion of Science (Y. Takashimizu, no. 24740341). We also acknowledge Dr. A Urabe (Niigata University) who assisted us in sampling.

Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.

Authors’ Affiliations

(1)
Faculty of Education, Niigata University
(2)
Oh-shima Elementary School

References

  1. Abramoff MD, Magalhaes PJ, Ram SJ (2004) Image processing with ImageJ. Biophotonics Int 11:36–42Google Scholar
  2. Arasan S, Akbulut S, Hasiloglu S (2011) The relationship between the fractal dimension and shape properties of particle. J Civil Eng 15:1219–25. doi:10.1007/s12205-011-1310-x Google Scholar
  3. Barrett PJ (1980) The shape of rock particles, a critical review. Sedimentology 27:291–303. doi:10.1111/j.1365-3091.1980.tb01179.x View ArticleGoogle Scholar
  4. Blott PJ, Pye K (2008) Particle shape: a review and new methods of characterization and classification. Sedimentology 55:31–63. doi:10.1111/j.1365-3091.2007.00892.x Google Scholar
  5. Bowman ET, Soga K, Drummond W (2001) Particle shape characterisation using Fourier descriptor analysis. Geotechnique 51:545–54View ArticleGoogle Scholar
  6. Clark MW (1981) Quantitative shape analysis: a review. Math Geol 13:303–20. doi:10.1007/BF01031516 View ArticleGoogle Scholar
  7. Cox EP (1927) A method of assigning numerical and percentage values to the degree of roundness of sand grains. J Paleontol 1:179–83Google Scholar
  8. Diepenbroek M, Bartholomä A, Ibbeken H (1992) How round is round? A new approach to the topic ‘roundness’ by Fourier grain shape analysis. Sedimentology 39:411–22. doi:10.1111/j.1365-3091.1992.tb02125.x View ArticleGoogle Scholar
  9. Drevin RG (2007) Computational methods for the determination of roundness of sedimentary particles. Math Geol 38:871–90. doi:10.1007/s11004-006-9051-y View ArticleGoogle Scholar
  10. ImageJ user guide, 2012. ImageJ/Fiji 1.46, p. 187.Google Scholar
  11. Itabashi K, Matsuo M, Naito M, Mori T (2004) Fractal analysis of visual chart of the particle shape and the comparison of shape parameters. Soils Found 44:143–56. doi:10.3208/sandf.44.143 (In Japanese)View ArticleGoogle Scholar
  12. Krumbein WC (1941) Measurement and geological significance of shape and roundness of sedimentary particles. J Sediment Petrol 11:64–72. doi:10.1306/D42690F3-2B26-11D7-8648000102C1865D View ArticleGoogle Scholar
  13. Lees G (1963) A new method determining the angularity of particles. Sedimentology 3:2–21. doi:10.1111/j.1365-3091.1964.tb00271.x View ArticleGoogle Scholar
  14. Lira C, Pina P (2009) Automated grain shape measurements applied to beach sands. J Coastal Res Spec Issue 56:1527–31Google Scholar
  15. Orford JD, Whalley WB (1983) The use of the fractal dimension to quantify the morphology of irregular-shaped particles. Sedimentology 30:655–68. doi:10.1111/j.1365-3091.1983.tb00700.x View ArticleGoogle Scholar
  16. Pettijohn FJ (1957) Sedimentary rocks, 2nd edn. Harper & Brothers, New YorkGoogle Scholar
  17. Powers MC (1953) A new roundness scale for sedimentary particles. J Sediment Petrol 23:117–9. doi:10.1306/D4269567-2B26-11D7-8648000102C1865D View ArticleGoogle Scholar
  18. Rittenhouse G (1943) A visual method of estimating two-dimensional sphericity. J Sediment Petrol 13:79–81Google Scholar
  19. Roussillon T, Piegay H, Sivignon I, Tougne L, Lavigne F (2009) Automatic computation of pebble roundness using digital imagery and discrete geometry. Comput Geosci 35:1992–2000. doi:10.1016/j.cageo.2009.01.013 View ArticleGoogle Scholar
  20. Schneider CA, Rasband WS, Eliceiri KW (2012) NIH Image to ImageJ: 25 years of image analysis. Nat Methods 9:671–5. doi:10.1038/nmeth.2089 View ArticleGoogle Scholar
  21. Schwarcz HP, Shane KC (1969) Measurements of particle shape by Fourier analysis. Sedimentology 13:213–31. doi:10.1111/j.1365-3091.1969.tb00170.x View ArticleGoogle Scholar
  22. Shimazu M, Tanaka K, Yoshida T (1970) Geology of the Taro district. Quadrangle series, scale 1:50,000, Akita no. 18, Geol Surv Japan. 54 (In Japanese with English abstract).Google Scholar
  23. Suzuki K, Sakai K, Ohta T (2013) Quantitative evaluation of grain shapes by utilizing Fourier and fractal analysis and implications for discriminating sedimentary environments. J Geol Soc Japan 119:205–16. doi:10.5575/geosoc.2012.0085 (In Japanese with English abstract)View ArticleGoogle Scholar
  24. Suzuki K, Fujiwara H, Ohta T (2015) The evaluation of macroscopic and microscopic textures of sand grains using elliptic Fourier and principal component analysis: implications for the discrimination of sedimentary environments. Sedimentology 62:1184–97. doi:10.1111/sed.12183 View ArticleGoogle Scholar
  25. Vallejo LE, Zhou Y (1995) The relationship between the fractal dimension and Krumbein’s roundness number. Soils Found 35:163–7. doi:10.3208/sandf1972.35.163 View ArticleGoogle Scholar
  26. Wadell H (1932) Volume, shape, and roundness of rock particles. J Geol 40:443–51View ArticleGoogle Scholar
  27. Winkelmolen AM (1982) Critical remarks on grain parameters, with special emphasis on shape. Sedimentology 29:255–65. doi:10.1111/j.1365-3091.1982.tb01722.x View ArticleGoogle Scholar
  28. Yoshimura Y, Ogawa S (1993) A simple quantification method of grain shape of granular materials such as sand. Doboku Gakkai Ronbunshu.J Japan Soc Civil Eng. No. 463/III-22:95–103 (In Japanese with English abstract). doi: 10.2208/jscej.1993.463_95

Copyright

© Takashimizu and Iiyoshi. 2016