Improved alignment of 3D FIB-SEM data for constructing digital rock nanoscale models
Iryna Reimers, Ilia Safonov, Ivan Yakimchuk
Studying porous materials at pore scale level is an important topic in general, and for the oil and gas industry particularly. Accurate data about internal structure of oil-bearing rocks helps to evaluate rock porosity and permeability, e.g. via digital rock technique. A digital rock model (DRM) — a digital representation of a sample — is one of the key elements in this approach. Generally, DRM is constructed relying on 3D image obtained with X-ray microtomography. In certain cases, images of much higher resolution are required. FIB-SEM technique can provide such images with resolved details at nanoscale.
FIB-SEM tomography produces a stack of images representing serial slices of the specimen. There are random displacements in both axes between adjacent slices. Therefore, an alignment procedure is required. Traditional alignment algorithms of a 3D image are based on comparison of two adjacent slices. These algorithms are easily confused by anisotropy in the porous sample structure or even geometry of experiment, i.e. they generate some additional erroneous shifts together with correction of “true” random displacements. In case of no spatially stable fiducial marks in the frame, such issue can notably distort the pore space geometry of constructed DRM. We present a new method, which meaningfully extends currently available alignment solutions.
Our approach allows to correct random shifts of slices and preserves the structure of the whole specimen at the same time. Displacements produced by traditional alignment algorithm are considered as a digital signal, e.g. dependence of horizontal shifts on slice number. Such signal can be decomposed on low and high-frequency components. Assuming that the FIB-SEM stack of images after preprocessing (e.g. curtaining effect removal) affected mainly by random shifts (that should be corrected), we suppress low frequency (slow) variations in displacement signal. Application of modified list of displacements results in much better alignment of FIB-SEM stack. The proposed algorithm can operate with arbitrary affine transformations (not only translations). Efficiency of described technique is demonstrated on synthetic and real images of rock samples.
