3
Image acquisition is the process of capturing a set of
overlapped video frames from a recorded borehole video
and image preprocessing techniques, such as grayscale con-
version and histogram equalization, can be carried out to
improve the key point detection. The video frames are cap-
tured, preprocessed, and stored as a list.
The next step is to estimate the homography between
every two adjacent frames in the video frame list by
detecting the key points in each frame and matching the
key points and estimating the geometric transformation
between every two adjacent frames. The process of key
point detection uses a feature detection algorithm to iden-
tify distinctive key points as features and extracts the associ-
ated descriptors. Different algorithms, e.g., Scale-Invariant
Feature Transform (SIFT), Speeded-Up Robust Features
(SURF), Oriented FAST and Rotated BRIEF (ORB), have
been used for feature detection in borehole images (Ma et
al. 2019 Zou et al. 2021a Deng et al. 2023)it is often
highly desirable to observe images of whole microscopic
sections with high resolution. So that micrograph stitching
is an important technology to produce a panorama or larger
image by combining multiple images with overlapping
areas, while retaining microscopic resolution. However,
due to high complexity and variety of microstructure, most
traditional methods could not balance speed and accuracy
of stitching strategy. To overcome this problem, we develop
a method named very fast sequential micrograph stitch-
ing (VFSMS. The ORB algorithm was found to have a
good performance in efficiency and accuracy for this study.
Examples of detected key points are present in Figure 2. For
each detected key point, one descriptor can be extracted
as a vector to represent the local appearance around each
detected key points, making it possible to match key points
across different images.
The goal of key point matching is to find correspon-
dence between the key points detected in two adjacent
images by matching the descriptors from one image to the
descriptors from another image. However, many key points
may have similar descriptors, especially in the regions
with repetitive patterns, and as a result, a key point in one
image may have multiple potential matches in another one,
as shown in Figure 3 where the potentially matched key
points are connected. It is essential to filter out ambiguous
matches to ensure robust and accurate matching. This is
normally done with a ratio test. Figure 4 shows the matched
key points after filtering with a ratio of 0.75, and significant
improvements in the overall accuracy of the feature match-
ing with the ratio test can be observed.
Offset calculation is the process to find the transforma-
tion between two adjacent frames based on the matched
key points. The transformations between two images
include translations, rotation, scaling, shearing, and per-
spective transformations. However, if a borescope moves in
a borehole and the distance from the borehole wall is main-
tained constant, the only transformation involved in the
recording process is translation when the borescope moves
vertically and rotates horizontally, presenting a considerable
scanning problem (Ma et al. 2019)it is often highly desir-
able to observe images of whole microscopic sections with
high resolution. So that micrograph stitching is an impor-
tant technology to produce a panorama or larger image by
combining multiple images with overlapping areas, while
retaining microscopic resolution. However, due to high
Figure 2. Key points that were detected with the ORB
algorithm
Figure 3. All matches between the detected key points in two
adjacent images
Figure 4. Filtered matches of key points between two images
Image acquisition is the process of capturing a set of
overlapped video frames from a recorded borehole video
and image preprocessing techniques, such as grayscale con-
version and histogram equalization, can be carried out to
improve the key point detection. The video frames are cap-
tured, preprocessed, and stored as a list.
The next step is to estimate the homography between
every two adjacent frames in the video frame list by
detecting the key points in each frame and matching the
key points and estimating the geometric transformation
between every two adjacent frames. The process of key
point detection uses a feature detection algorithm to iden-
tify distinctive key points as features and extracts the associ-
ated descriptors. Different algorithms, e.g., Scale-Invariant
Feature Transform (SIFT), Speeded-Up Robust Features
(SURF), Oriented FAST and Rotated BRIEF (ORB), have
been used for feature detection in borehole images (Ma et
al. 2019 Zou et al. 2021a Deng et al. 2023)it is often
highly desirable to observe images of whole microscopic
sections with high resolution. So that micrograph stitching
is an important technology to produce a panorama or larger
image by combining multiple images with overlapping
areas, while retaining microscopic resolution. However,
due to high complexity and variety of microstructure, most
traditional methods could not balance speed and accuracy
of stitching strategy. To overcome this problem, we develop
a method named very fast sequential micrograph stitch-
ing (VFSMS. The ORB algorithm was found to have a
good performance in efficiency and accuracy for this study.
Examples of detected key points are present in Figure 2. For
each detected key point, one descriptor can be extracted
as a vector to represent the local appearance around each
detected key points, making it possible to match key points
across different images.
The goal of key point matching is to find correspon-
dence between the key points detected in two adjacent
images by matching the descriptors from one image to the
descriptors from another image. However, many key points
may have similar descriptors, especially in the regions
with repetitive patterns, and as a result, a key point in one
image may have multiple potential matches in another one,
as shown in Figure 3 where the potentially matched key
points are connected. It is essential to filter out ambiguous
matches to ensure robust and accurate matching. This is
normally done with a ratio test. Figure 4 shows the matched
key points after filtering with a ratio of 0.75, and significant
improvements in the overall accuracy of the feature match-
ing with the ratio test can be observed.
Offset calculation is the process to find the transforma-
tion between two adjacent frames based on the matched
key points. The transformations between two images
include translations, rotation, scaling, shearing, and per-
spective transformations. However, if a borescope moves in
a borehole and the distance from the borehole wall is main-
tained constant, the only transformation involved in the
recording process is translation when the borescope moves
vertically and rotates horizontally, presenting a considerable
scanning problem (Ma et al. 2019)it is often highly desir-
able to observe images of whole microscopic sections with
high resolution. So that micrograph stitching is an impor-
tant technology to produce a panorama or larger image by
combining multiple images with overlapping areas, while
retaining microscopic resolution. However, due to high
Figure 2. Key points that were detected with the ORB
algorithm
Figure 3. All matches between the detected key points in two
adjacent images
Figure 4. Filtered matches of key points between two images