Stitching a series of geologic core images

[GeologyAMG]

Hello,

I am working with geologic core photos and I am attempting to mimic a line scanner with my ipad. I collect a series of sequential photos (left to right along the length of a core). The photos should have >50% overlap. However when I attempt to use OpenStitching to combine these photos the result is often a truncated portion of the entire length. Could this be due to the files sizes or a stitching setting? I am happy to share the original data etc. if someone can assist. Files are 55Mb zipped and won't add to this. My code is in the .ipynb file, just remove the .txt extension.

4_Stitch.ipynb.txt

[lukasalexanderweber]

can you first try verbose stitching to see e.g. the resulting matches and confidence matrices?

[GeologyAMG]

When I ran verbose stitching I get this as an output, however the images is still truncated.

Path to the folder containing the images

img_dir = r"test_images/original images"
output_file = r"test_images/stitched_output.jpg"

Get all jpg files and sort them by filename

img_paths = sorted(glob.glob(os.path.join(img_dir, "*.jpg")))
print(f"Found {len(img_paths)} images:")
for path in img_paths:
print(f" {path}")

Read images

images = []
for img_path in img_paths:
img = cv2.imread(img_path)
if img is not None:
print(f"Loaded {img_path} with shape {img.shape}")
images.append(img)
else:
print(f"Warning: Could not load {img_path}")

if len(images) < 2:
raise Exception("Need at least two images to stitch.")

Create stitcher and stitch images

try:
stitcher = cv2.Stitcher_create()
except AttributeError:
stitcher = cv2.createStitcher()

print("Stitching images...")
status, stitched = stitcher.stitch(images)

if status == cv2.Stitcher_OK:
cv2.imwrite(output_file, stitched)
print(f"Stitched image saved as {output_file}")
else:
print(f"Stitching failed with status {status}")

Found 32 images:
test_images/original images\1743083839698654.jpg
test_images/original images\1743083841406210.jpg
test_images/original images\1743083843173261.jpg
test_images/original images\1743083845049521.jpg
test_images/original images\1743083845885234.jpg
test_images/original images\1743083847432958.jpg
test_images/original images\1743083849009921.jpg
test_images/original images\1743083850593792.jpg
test_images/original images\1743083852501247.jpg
test_images/original images\1743083854151743.jpg
test_images/original images\1743083855647681.jpg
test_images/original images\1743083857546742.jpg
test_images/original images\1743083859163638.jpg
test_images/original images\1743083860613251.jpg
test_images/original images\1743083862649780.jpg
test_images/original images\1743083863874680.jpg
test_images/original images\1743083865342795.jpg
test_images/original images\1743083867117415.jpg
test_images/original images\1743083868597699.jpg
test_images/original images\1743083870545863.jpg
test_images/original images\1743083872026218.jpg
test_images/original images\1743083873421025.jpg
test_images/original images\1743083874798462.jpg
test_images/original images\1743083876835345.jpg
test_images/original images\1743083878143405.jpg
test_images/original images\1743083879480876.jpg
test_images/original images\1743083880818869.jpg
test_images/original images\1743083881977302.jpg
test_images/original images\1743083883662974.jpg
test_images/original images\1743083885071042.jpg
test_images/original images\1743083886411846.jpg
test_images/original images\1743083887725190.jpg
Loaded test_images/original images\1743083839698654.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083841406210.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083843173261.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083845049521.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083845885234.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083847432958.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083849009921.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083850593792.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083852501247.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083854151743.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083855647681.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083857546742.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083859163638.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083860613251.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083862649780.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083863874680.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083865342795.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083867117415.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083868597699.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083870545863.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083872026218.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083873421025.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083874798462.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083876835345.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083878143405.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083879480876.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083880818869.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083881977302.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083883662974.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083885071042.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083886411846.jpg with shape (2124, 2832, 3)
Loaded test_images/original images\1743083887725190.jpg with shape (2124, 2832, 3)
Stitching images...
Stitched image saved as test_images/stitched_output.jpg

[GeologyAMG]

It looks like it finds matches for all the photos.

img_dir = r"test_images/original images"
img_paths = sorted(glob.glob(os.path.join(img_dir, "*.jpg")))

Read images

images = []
for img_path in img_paths:
img = cv2.imread(img_path)
if img is not None:
images.append(img)

if len(images) < 2:
raise Exception("Need at least two images to match.")

orb = cv2.ORB_create()

for i in range(len(images) - 1):
img1 = images[i]
img2 = images[i + 1]
kp1, des1 = orb.detectAndCompute(img1, None)
kp2, des2 = orb.detectAndCompute(img2, None)
print(f"Image {i} vs {i+1}:")
print(f" Keypoints in image {i}: {len(kp1)}")
print(f" Keypoints in image {i+1}: {len(kp2)}")
if des1 is not None and des2 is not None:
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
matches = bf.match(des1, des2)
print(f" Matches: {len(matches)}")
else:
print(" Warning: Could not compute descriptors for one or both images.")

Image 0 vs 1:
Keypoints in image 0: 500
Keypoints in image 1: 500
Matches: 183
Image 1 vs 2:
Keypoints in image 1: 500
Keypoints in image 2: 500
Matches: 184
Image 2 vs 3:
Keypoints in image 2: 500
Keypoints in image 3: 500
Matches: 174
Image 3 vs 4:
Keypoints in image 3: 500
Keypoints in image 4: 500
Matches: 269
Image 4 vs 5:
Keypoints in image 4: 500
Keypoints in image 5: 500
Matches: 176
Image 5 vs 6:
Keypoints in image 5: 500
Keypoints in image 6: 500
Matches: 186
Image 6 vs 7:
Keypoints in image 6: 500
Keypoints in image 7: 500
Matches: 178
Image 7 vs 8:
Keypoints in image 7: 500
Keypoints in image 8: 500
Matches: 189
Image 8 vs 9:
Keypoints in image 8: 500
Keypoints in image 9: 500
Matches: 180
Image 9 vs 10:
Keypoints in image 9: 500
Keypoints in image 10: 500
Matches: 186
Image 10 vs 11:
Keypoints in image 10: 500
Keypoints in image 11: 500
Matches: 164
Image 11 vs 12:
Keypoints in image 11: 500
Keypoints in image 12: 500
Matches: 170
Image 12 vs 13:
Keypoints in image 12: 500
Keypoints in image 13: 500
Matches: 173
Image 13 vs 14:
Keypoints in image 13: 500
Keypoints in image 14: 500
Matches: 176
Image 14 vs 15:
Keypoints in image 14: 500
Keypoints in image 15: 500
Matches: 190
Image 15 vs 16:
Keypoints in image 15: 500
Keypoints in image 16: 500
Matches: 236
Image 16 vs 17:
Keypoints in image 16: 500
Keypoints in image 17: 500
Matches: 159
Image 17 vs 18:
Keypoints in image 17: 500
Keypoints in image 18: 500
Matches: 175
Image 18 vs 19:
Keypoints in image 18: 500
Keypoints in image 19: 500
Matches: 186
Image 19 vs 20:
Keypoints in image 19: 500
Keypoints in image 20: 500
Matches: 176
Image 20 vs 21:
Keypoints in image 20: 500
Keypoints in image 21: 500
Matches: 164
Image 21 vs 22:
Keypoints in image 21: 500
Keypoints in image 22: 500
Matches: 173
Image 22 vs 23:
Keypoints in image 22: 500
Keypoints in image 23: 500
Matches: 148
Image 23 vs 24:
Keypoints in image 23: 500
Keypoints in image 24: 500
Matches: 162
Image 24 vs 25:
Keypoints in image 24: 500
Keypoints in image 25: 500
Matches: 166
Image 25 vs 26:
Keypoints in image 25: 500
Keypoints in image 26: 500
Matches: 165
Image 26 vs 27:
Keypoints in image 26: 500
Keypoints in image 27: 500
Matches: 198
Image 27 vs 28:
Keypoints in image 27: 500
Keypoints in image 28: 500
Matches: 158
Image 28 vs 29:
Keypoints in image 28: 500
Keypoints in image 29: 500
Matches: 169
Image 29 vs 30:
Keypoints in image 29: 500
Keypoints in image 30: 500
Matches: 182
Image 30 vs 31:
Keypoints in image 30: 500
Keypoints in image 31: 500
Matches: 185

[GeologyAMG]

I am using your Stitching Tutorial.ipynb and brought in my 32 files. It is still 50MB so I can't share the whole thing. Here is the subset output.

[lukasalexanderweber]

Thanks! Two things regarding your use case:

Use a RangeWidthMatcher with range width 1. This ensures, that only Images which are neighbours are tried to Stitch.

Set the confidence threshold to 0 as you know that those neighbours definetly match

[GeologyAMG]

Thanks, I tried to apply with the following:

Load images from file paths if needed

core_img_arrays = [cv.imread(path) for path in core_imgs]

orb = cv.ORB_create()
bf = cv.BFMatcher(cv.NORM_HAMMING, crossCheck=True)
confidence_threshold = 0 # Accept all matches

Compute keypoints and descriptors

kp_des = [orb.detectAndCompute(img, None) for img in core_img_arrays]

Only match adjacent images (i, i+1)

matches_list = []
for i in range(len(core_imgs) - 1):
des1 = kp_des[i][1]
des2 = kp_des[i + 1][1]
if des1 is not None and des2 is not None:
matches = bf.match(des1, des2)
# Filter matches by confidence threshold (distance)
good_matches = [m for m in matches if m.distance >= confidence_threshold]
matches_list.append(good_matches)
print(f"Image {i} <-> {i+1}: {len(good_matches)} matches")
else:
matches_list.append([])
print(f"Image {i} <-> {i+1}: No descriptors found.")

Use OpenCV's Stitcher (it will use its own matcher, but this shows how to restrict matches)

try:
stitcher = cv.Stitcher_create()
except AttributeError:
stitcher = cv.createStitcher()

stitcher.setPanoConfidenceThresh(confidence_threshold)
status, stitched = stitcher.stitch(core_img_arrays)

if status == cv.Stitcher_OK:
cv.imwrite("stitched_core_img.jpg", stitched)
print("Stitched image saved as stitched_core_img.jpg")
else:
print(f"Stitching failed with status {status}")

Image 0 <-> 1: 163 matches
Image 1 <-> 2: 186 matches
Image 2 <-> 3: 164 matches
Image 3 <-> 4: 170 matches
Image 4 <-> 5: 173 matches
Image 5 <-> 6: 176 matches
Image 6 <-> 7: 190 matches
Image 7 <-> 8: 236 matches
Image 8 <-> 9: 159 matches
Image 9 <-> 10: 175 matches
Image 10 <-> 11: 160 matches
Image 11 <-> 12: 160 matches
Image 12 <-> 13: 176 matches
Image 13 <-> 14: 164 matches
Image 14 <-> 15: 173 matches
Image 15 <-> 16: 148 matches
Image 16 <-> 17: 162 matches
Image 17 <-> 18: 166 matches
Image 18 <-> 19: 165 matches
Image 19 <-> 20: 198 matches
Image 20 <-> 21: 158 matches
Image 21 <-> 22: 159 matches
Image 22 <-> 23: 148 matches
Image 23 <-> 24: 182 matches
Image 24 <-> 25: 185 matches
Image 25 <-> 26: 154 matches
Image 26 <-> 27: 269 matches
Image 27 <-> 28: 176 matches
Image 28 <-> 29: 186 matches
Image 29 <-> 30: 178 matches
Image 30 <-> 31: 189 matches
Stitching failed with status 3