Pore Backs Segmentation in PerGeos

Pore-backs segmentation in PerGeos involves the following steps: 1. Using a variance filter to detect flat regions of the rock matrix and avoid marking artifacts in pore-to-matrix slopes. 2. Applying thresholding to add markers in the matrix and pores. 3. Performing a 2D gradient-based Watershed Transform on the merged markers to segment pore-backs. 4. An alternative workflow first thresholds to mark pores and matrix, then uses a small variance kernel and closing to remove noise before Watershed segmentation.

Uploaded by

downbuliao

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

45 views

Pore Backs Segmentation in PerGeos

Uploaded by

downbuliao

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 6

Pore-backs segmentation in PerGeos

Introduction

Imagery produced from FIBSEMs (Dualbeams) provides users with unique data in terms of resolution and sample size
that are not available with other techniques. Despite the power of this analytical approach, some limitations in
quantifying the textural data restrict the accuracy of data obtained from FIBSEM images.

These limitations are derived from artifacts that occur during the acquisition of the data. Collecting data using a FIBSEM
is a destructive, serial sectioning technique. A thin slice of material is milled away, the resulting surface imaged, and the
process repeated until the requested volume is generated. However, porous materials, by their very nature, create some
issues in how to interpret the data that are produced. When each slice is imaged, the empty space (pores and large
cracks) are included in the images. If the pore is not deep enough to trap electrons forming a “black” pore synonymous
with empty space in SEM imaging, or if it has been filled with redeposited material as a result of the milling process, the
back of the pore or other pore filling materials will be imaged. These “pore backs” often termed “pore shine” or “shine
through” artifacts make defining the true area of a pore in 2D (volume in 3D) extremely difficult to quantify during post
processing using image analysis software.

While ubiquitous, particularly in tight rocks and carbonates, these artifacts can be accounted for. This communication
describes some workflows and approaches available using the PerGeos software for Digital Rock analysis. The workflows
presented are easy to replicate and automate using PerGeos’ unique automation and processing capabilities and
libraries. The workflows presented below are valid for 2D and 3D SEM/FIBSEM data.
Figure 1 Highlighted pore back in a FIB/SEM slice

Methodology

The pore-back segmentation will use a 2D expansion by a Watershed Transform from markers.

A 2D rather than 3D watershed is preferred in the case of FIB/SEM images, since the slice spacing is generally much
bigger than the pixel size. The difference between 2 consecutive slices creates a high gradient in the Z direction, blocking
the usage of a 3D Watershed.

If we use a threshold-based approach to add markers in the matrix, artefacts will appear in the pores-to-matrix slopes,
since these regions share the same greyscale intensities:

Figure 2 High value intensity thresholdingfor tentatively marking the matrix

The global idea of our method is to detect flat regions of the matrix with a non-local deviation filter, the variance filter,
which unlike the gradient will not have a high intensity in the pore-to-matrix slopes. This way, marking artefacts are
avoided, and the high intensity around the pores will make a Watershed Transform on a 2D gradient very effective.

The re-deposition or charging around the pores will even make the process stronger. A very high thresholding to enhance
pore markers is then an interesting option.
Figure 3 Re-deposition around the pores

In addition to this intelligent workflow, the recipe mechanism of PerGeos makes it easily accessible to any user,
repeatable and applicable with light modifications to any data.

The PerGeos recipe incorporates the following steps:

Compute the variance filter ( kernel size of 20-30 )

Mark the matrix with a low variance thresholding

Mark the pores ( done in purpose after the matrix to override low variance pores detected in the previous step ) with low
thresholding + optional black tophat

Enhance the pores with a high value thresholding ( optional )

Apply a 2D gradient-based Watershed Transform

Enhance the pores with a black TopHat

Figure 4 PerGeos Pore back recipe

The recipe into details

Figure 5 Greyscale image Figure 6 Variance filter (kernel size of 20)

Figure 7 Matrix markers Figure 8 Pore markers after low thresholding

Figure 9 Merged markers Figure 10 2D Gradient-based Watershed

Figure 11 Enhancing the pores with a black topHat

Alternatives

As for many other segmentation workflows, some alternatives are sometimes interesting.

The following workflow has also proven its efficiency on some FIB/SEM data. The major difference is due to the
heterogeneity of the matrix (mineral inclusions, organic matter all around the pores) that makes the use of a large
variance kernel not possible.

On the opposite, a threshold in the pores is only catching some noise in the pore-to-matrix slopes.

This way, we can use a thresholding to mark both the pores and matrix (enhanced by a topHat), and use a variance with a
small kernel, made bigger by a closing, to remove the noise or false markers.

Then, the 2D Watershed expands the markers and catches the pore-backs.
This PerGeos alternative recipe incorporates the following steps: