Academia.eduAcademia.edu

Figure 1 – uploaded by Parnian Kasae

See full PDFdownloadDownload figure
Fig. 1. Schema of the Pore3D software architecture which emphasizes the different ways to take advantage of Pore3D functionalities.

Figure 1 Schema of the Pore3D software architecture which emphasizes the different ways to take advantage of Pore3D functionalities.

Related Figures (9)

ig. 2. Analysis of an alginate/hydroxyapatite composite scaffold sample (material courtesy of G. Turco, Department of Life Sciences, University of Trieste, Italy): (a) X-ray —CT slice imaged with a conventional source (V=40 kV, [= 200A, focal spot = 8 um, voxel size: 6.25 ttm); (b) schema of the proposed image analysis protocol witl lated computational time and results (computational time recorded on a Intel Core 2 Duo E8200 2.66 GHz machine). “In this section, the analysis of a milk aerated chocolate sample is reported [38]. The quality of chocolate is related to its sensorial In addition to the morphometric parameters, this example includes the box counting fractal dimension D as a textural descriptor which is a dimensionless value that lies within the range of 2-3. Its value is close to 2 for smooth surfaces and higher values for increasing surface roughness or complexity [32]. The
ig. 2. Analysis of an alginate/hydroxyapatite composite scaffold sample (material courtesy of G. Turco, Department of Life Sciences, University of Trieste, Italy): (a) X-ray —CT slice imaged with a conventional source (V=40 kV, [= 200A, focal spot = 8 um, voxel size: 6.25 ttm); (b) schema of the proposed image analysis protocol witl lated computational time and results (computational time recorded on a Intel Core 2 Duo E8200 2.66 GHz machine). “In this section, the analysis of a milk aerated chocolate sample is reported [38]. The quality of chocolate is related to its sensorial In addition to the morphometric parameters, this example includes the box counting fractal dimension D as a textural descriptor which is a dimensionless value that lies within the range of 2-3. Its value is close to 2 for smooth surfaces and higher values for increasing surface roughness or complexity [32]. The
Fig. 3. Analysis of an alginate/hydroxyapatite composite scaffold sample (material courtesy of G. Turco, Department of Life Sciences, University of Trieste, Italy): (a) original image; (b) thresholded image; (c) volume rendering of the 400? voxels REV. (rendering performed using the commercial software VGStudio MAX 2.0).
Fig. 3. Analysis of an alginate/hydroxyapatite composite scaffold sample (material courtesy of G. Turco, Department of Life Sciences, University of Trieste, Italy): (a) original image; (b) thresholded image; (c) volume rendering of the 400? voxels REV. (rendering performed using the commercial software VGStudio MAX 2.0).
Fig. 4. Analysis of a milk aerated chocolate sample: (a) monochromatic synchrotron X-ray 1—CT slice (energy = 13keV, voxel size =14|1m, sample-to-detector distance = 50mm); (b) schema of the proposed image analysis protocol with related computational time and results (computational time recorded on a Intel Core 2 Duo E8200 2.66 GHz machine). = For this example, a Volume Of Interest (VOI) of 200? voxels was extracted from the original data set (a representative slice of the VOI is reported in Fig. 5a) and segmented by means of a bilateral filtering [40] followed by Otsu’s thresholding [27] (see Figs. 5b and c). Visually inspecting the segmented VOI it can be noticed that some bubbles are interconnected, e.g. some visually distinguishable bubbles form a connected component. To empha- size this situation, the connected components labeled image has been represented using a random color look up table (Fig. 5d) as commonly used. Due to the convexity of the bubbles of this chocolate sample it is quite easy to perform the separation of “touching” bubbles with few morphological image processing steps. The chessboard distance transform [41] of the solid phase was computed and its intensity was inverted (Fig. 5e). Subse- quentially, the watershed transformation [42] was computed resulting in a network of lines that segment the image into disjoint regions, each including an individual bubble (Fig. 5f). Masking this watershed segmented image with the thresholded image a binary image is obtained in which each bubble has been identified and separated (Figs. 5g and h). Performing the connected components labeling of this image it is possible to
Fig. 4. Analysis of a milk aerated chocolate sample: (a) monochromatic synchrotron X-ray 1—CT slice (energy = 13keV, voxel size =14|1m, sample-to-detector distance = 50mm); (b) schema of the proposed image analysis protocol with related computational time and results (computational time recorded on a Intel Core 2 Duo E8200 2.66 GHz machine). = For this example, a Volume Of Interest (VOI) of 200? voxels was extracted from the original data set (a representative slice of the VOI is reported in Fig. 5a) and segmented by means of a bilateral filtering [40] followed by Otsu’s thresholding [27] (see Figs. 5b and c). Visually inspecting the segmented VOI it can be noticed that some bubbles are interconnected, e.g. some visually distinguishable bubbles form a connected component. To empha- size this situation, the connected components labeled image has been represented using a random color look up table (Fig. 5d) as commonly used. Due to the convexity of the bubbles of this chocolate sample it is quite easy to perform the separation of “touching” bubbles with few morphological image processing steps. The chessboard distance transform [41] of the solid phase was computed and its intensity was inverted (Fig. 5e). Subse- quentially, the watershed transformation [42] was computed resulting in a network of lines that segment the image into disjoint regions, each including an individual bubble (Fig. 5f). Masking this watershed segmented image with the thresholded image a binary image is obtained in which each bubble has been identified and separated (Figs. 5g and h). Performing the connected components labeling of this image it is possible to
Fig. 5. Steps required to achieve a correct segmentation and isolation of pores in the milk aerated chocolate sample. For the sake of clarity only one slice of the 200? voxels VOI (voxel size =14,1m) is represented: (a) original image; (b) filtered image with bilateral filter (¢, = 3.0, og =30.0); (c) thresholded image with Otsu’s method (automatically proposed threshold !=65); (d) connected components labeling of image c (393 connected components found); (e) chessboard distance transform of the background of image c with intensity inverted (the intensity has been rescaled for representation purposes); (f) watershed segmentation (binary image with watershed lines) of image e; (g) masking (logical sum of white voxels) of image c and image f; (h) connected components labeling of image f (583 connected components found). The separation process is evident comparing images d and h.
Fig. 5. Steps required to achieve a correct segmentation and isolation of pores in the milk aerated chocolate sample. For the sake of clarity only one slice of the 200? voxels VOI (voxel size =14,1m) is represented: (a) original image; (b) filtered image with bilateral filter (¢, = 3.0, og =30.0); (c) thresholded image with Otsu’s method (automatically proposed threshold !=65); (d) connected components labeling of image c (393 connected components found); (e) chessboard distance transform of the background of image c with intensity inverted (the intensity has been rescaled for representation purposes); (f) watershed segmentation (binary image with watershed lines) of image e; (g) masking (logical sum of white voxels) of image c and image f; (h) connected components labeling of image f (583 connected components found). The separation process is evident comparing images d and h.
Fig. 6. Volume rendering of the 200° voxels VOI (voxel size = 141m) for the milk aerated chocolate sample in which each pore is represented with a different label: (a) whole VOI (583 connected components); (b) REV without “incomplete” pores, i.e. pores “touching” VOI margins (339 connected components remaining). (rendering performed using the commercial software VGStudio MAX 2.0). The skeleton (or medial axis) is an effective tool for the analysis of porous materials [43]. The skeleton of any 3D object is intuitively the “spine” of the object running along its geometric middle. For example, the medial axis for a sphere is the center point and for a cylinder it is the axis of rotational symmetry. By scanning the skeleton it is possible to generate the equivalent graph of nodes and branches, thus allowing a_ topological characterization of the underlying object. It is also possible to add geometrical measurements like the length and the width of the branches. For the latter measure, it suffices to take advantage of the distance transform of the object (this is related to the maximal spheres method [44]). While being a general tool, the skeleton analysis is practically used for objects presenting a tubular structure or having an interconnected void space where channels or paths may be recognized. For this kind of porous media parameters like interconnectivity and geometrical tortu- osity are of great interest [43]. The effectiveness of the skeleton analysis is limited by the accuracy of the skeletonization algorithm. One of the difficulties is that a skeleton (or, more precisely a “curve-skeleton”) is an ill-defined object. This has led to a large number of algorithms in the literature and for some of them it is unclear how general and robust they are [45]. In practical cases, the skeletonization process requires some sort of pre- and post-processing. It could be useful to apply the actual skeletonization algorithm on a binary smoothed version of the segmented image in order to hopefully reduce the amount of spurious branches and also a pruning post-processing is often adopted in order to improve the quality of the result. Pore3D offers several skeletonization algorithms [46-48] as well as pruning methods [49] allowing to compute a reliable skeleton in most practical cases.
Fig. 6. Volume rendering of the 200° voxels VOI (voxel size = 141m) for the milk aerated chocolate sample in which each pore is represented with a different label: (a) whole VOI (583 connected components); (b) REV without “incomplete” pores, i.e. pores “touching” VOI margins (339 connected components remaining). (rendering performed using the commercial software VGStudio MAX 2.0). The skeleton (or medial axis) is an effective tool for the analysis of porous materials [43]. The skeleton of any 3D object is intuitively the “spine” of the object running along its geometric middle. For example, the medial axis for a sphere is the center point and for a cylinder it is the axis of rotational symmetry. By scanning the skeleton it is possible to generate the equivalent graph of nodes and branches, thus allowing a_ topological characterization of the underlying object. It is also possible to add geometrical measurements like the length and the width of the branches. For the latter measure, it suffices to take advantage of the distance transform of the object (this is related to the maximal spheres method [44]). While being a general tool, the skeleton analysis is practically used for objects presenting a tubular structure or having an interconnected void space where channels or paths may be recognized. For this kind of porous media parameters like interconnectivity and geometrical tortu- osity are of great interest [43]. The effectiveness of the skeleton analysis is limited by the accuracy of the skeletonization algorithm. One of the difficulties is that a skeleton (or, more precisely a “curve-skeleton”) is an ill-defined object. This has led to a large number of algorithms in the literature and for some of them it is unclear how general and robust they are [45]. In practical cases, the skeletonization process requires some sort of pre- and post-processing. It could be useful to apply the actual skeletonization algorithm on a binary smoothed version of the segmented image in order to hopefully reduce the amount of spurious branches and also a pruning post-processing is often adopted in order to improve the quality of the result. Pore3D offers several skeletonization algorithms [46-48] as well as pruning methods [49] allowing to compute a reliable skeleton in most practical cases.
Fig. 7. Analysis of the porous network created by a boring sponge Cliona celata in a limestone pebble. From left to right: (a) X-ray u—CT slice imaged using a conventional source (V=130kV, I= 123 pA, focal spot = 201m, voxel size = 19.4 1m); (b) schema of the proposed image analysis protocol with related computational time and results (computational time recorded on a Intel Core 2 Duo E8200 2.66 GHz machine).
Fig. 7. Analysis of the porous network created by a boring sponge Cliona celata in a limestone pebble. From left to right: (a) X-ray u—CT slice imaged using a conventional source (V=130kV, I= 123 pA, focal spot = 201m, voxel size = 19.4 1m); (b) schema of the proposed image analysis protocol with related computational time and results (computational time recorded on a Intel Core 2 Duo E8200 2.66 GHz machine).
Fig. 8. Steps required to perform an effective skeleton analysis of the porous network created by a boring sponge Cliona celata in a limestone pebble. For the sake of clar only one slice of the 400 x 400 x 100 voxels VOI (voxel size = 19.5 jim) is reported: (a) original image; (b) result of ring remover filtering [50] applied on image a; (c) res! of nonlinear diffusion [51] applied on image b; (d) thresholding of image c using a fixed threshold automatically determined according to Johannsen’s method [52]. As an example, the skeletonization algorithm has been applied to a u—CT data set of a boring sponge Cliona celata in a limestone pebble imaged at the TOMOLAB station. In Fig. 7a a reconstructed slice of this sample is shown together with some parameters extracted by quantitative analysis (Fig. 7b). The study of the imestone fragment housing the boring sponge can provide information about the growth mechanisms of the sponge, with particular attention to the size and connectivity of the burrows. Due to the presence of severe pronounced ring artifacts, images have been pre-processed (Fig. 8) by means of a ring artifacts reduction filter custom-implemented [50] in the Pore3D software. In order to optimize the following segmentation step, few iterations of a nonlinear diffusion filter [51] have been applied. The segmentation has been performed with the automatic thresholding method proposed by Johannsen [52]. Although the segmented image is a valid input for the skeletonization process, the skeletonization algorithm proposed in [46] has been applied on a binary median smoothed version of the same image In the scientific community there is a strong interest in multiphase materials, where the phases correspond to compo- nents with different chemical composition and/or density and can correspond to different gray levels in the X-ray tomograms. In this case Pore3D gives the possibility to separate each phase implementing 3D clustering (k-means, k-medians) and several
Fig. 8. Steps required to perform an effective skeleton analysis of the porous network created by a boring sponge Cliona celata in a limestone pebble. For the sake of clar only one slice of the 400 x 400 x 100 voxels VOI (voxel size = 19.5 jim) is reported: (a) original image; (b) result of ring remover filtering [50] applied on image a; (c) res! of nonlinear diffusion [51] applied on image b; (d) thresholding of image c using a fixed threshold automatically determined according to Johannsen’s method [52]. As an example, the skeletonization algorithm has been applied to a u—CT data set of a boring sponge Cliona celata in a limestone pebble imaged at the TOMOLAB station. In Fig. 7a a reconstructed slice of this sample is shown together with some parameters extracted by quantitative analysis (Fig. 7b). The study of the imestone fragment housing the boring sponge can provide information about the growth mechanisms of the sponge, with particular attention to the size and connectivity of the burrows. Due to the presence of severe pronounced ring artifacts, images have been pre-processed (Fig. 8) by means of a ring artifacts reduction filter custom-implemented [50] in the Pore3D software. In order to optimize the following segmentation step, few iterations of a nonlinear diffusion filter [51] have been applied. The segmentation has been performed with the automatic thresholding method proposed by Johannsen [52]. Although the segmented image is a valid input for the skeletonization process, the skeletonization algorithm proposed in [46] has been applied on a binary median smoothed version of the same image In the scientific community there is a strong interest in multiphase materials, where the phases correspond to compo- nents with different chemical composition and/or density and can correspond to different gray levels in the X-ray tomograms. In this case Pore3D gives the possibility to separate each phase implementing 3D clustering (k-means, k-medians) and several
Fig. 9. Rendering of a 7.80 x 7.80 x 1.95 mm? VOI and related skeleton of the porous network created by the boring sponge Cliona celata in a limestone pebble. By scanning the skeleton it is possible to extract topological features thus allowing a characterization of the connectivity and tortuosity of the porous network (rendering performed using the commercial software VGStudio MAX 2.0).
Fig. 9. Rendering of a 7.80 x 7.80 x 1.95 mm? VOI and related skeleton of the porous network created by the boring sponge Cliona celata in a limestone pebble. By scanning the skeleton it is possible to extract topological features thus allowing a characterization of the connectivity and tortuosity of the porous network (rendering performed using the commercial software VGStudio MAX 2.0).
Fig. 10. Scoria sample from the Villarrica volcano (Chile) imaged by means of monochromatic synchrotron radiation 11—CT (energy = 28 keV, voxel size = 14|1m, sample-to-detector distance = 130mm): (a) reconstructed slice; (b) different phases represented using different labels.
Fig. 10. Scoria sample from the Villarrica volcano (Chile) imaged by means of monochromatic synchrotron radiation 11—CT (energy = 28 keV, voxel size = 14|1m, sample-to-detector distance = 130mm): (a) reconstructed slice; (b) different phases represented using different labels.
Related topics:
Computer SciencePhysicsImage ProcessingSoftware3-D ImagingX ray Computed TomographySoftware Librarycomputed microtomography
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved