Otolith Image Processing Library (OIPL)

Input Image

Results

Details (click here for more information)

Below are the 9 closest matches from our Image Library to your unknown image. Provided are the Family and Species name of each match, the distance score between the unknown and the matched image (smaller is better), and the file name of each matched image.

A green distance score and border indicates that during cross-validation, a match with this distance score was always the correct species.
A yellow distance score and border indicates a cross-validation correct match to Genus but not necessarily species.
An orange distance score and border indicates that in cross validation this score was always the correct Family.

Click on the Species name or image to go to the Statistics page for that species, and browse all library images and cross validation results.
Please note: the OIPL will always provide the 9 best matches. This does not mean any are correct, even if they are colored green based on the distance score, they are simply the 9 closest images in our library to your unknown. Users should use their judgement to decide on a final identification.

History

Click on a previously uploaded image link below to re-load the results. Note that these results are only kept for your current browsing session.

Download Results Clear History

The OIPL will assist researchers in identifying fish remains based on images of their otoliths. It is primarily aimed at researchers conducting stomach content analysis, to identify partially digested fish remains from the otoliths, but may also be of use to archaeologists and paleontologists.

To use, simply “Choose File” above to navigate and select an image file of an unknown otolith (JPG, BMP, TIF, PNG), and “Upload Image”. The machine learning algorithms we developed will then compare your unknown image to our library and return the best 9 matches. The OIPL currently contains over 4200 images from 102 taxa of common estuarine and inshore fishes of the central northern Gulf of Mexico. Users can browse the full image library to evaluate the results. Note, this is a tool to assist researchers. The closest match is not always correct, every image you upload will generate matches, but your species may not be represented in our library. You can see more by browsing the statistics.

We have developed a simple Guide to Otolith Imaging to assist users in producing images suitable for analysis by the OIPL. The tool will build a table of results, collating all outputs from each session, allowing users to upload and identify multiple images in a session, and then export the results.

Model Validation

Internal cross-validation of the model involves feeding each image from the library into the model as an unknown, and extracting the distance score between each image and every other image in the library. The Classification Accuracy on the Statistics page and Species list reflect the percentage of images in that class for which the closest match was to an image from the same class (image analytical class, not taxonomic Class: most classes correspond to Species). The current model and library has an overall cross-validation accuracy of over 93%, meaning that when the 4200 images in the library are fed into the model as unknowns, for over 93% the closest match is another image from the same class. Few classes with over 20 images have a cross-validation accuracy below 90%.

In the Results output of the 9 closest matches, a green distance score and border indicates a distance that in cross-validation for that class was always a correct match, or, that no images from other classes in the library matched to any image in the presented class with a smaller distance score. A yellow distance score and border indicates that cross-validation matches with this distance score were always the correct Genus, if not always the correct species. Orange distance score and border indicates cross-validation scores that were always correct to Family for the presented class.

Further validation was undertaken using a collection of otoliths recovered from the stomachs of red drum and spotted seatrout from coastal Alabama. Based on our subjective identification of the unknown otoliths, we considered that ~500 were represented in the OIPL, and that it correctly identified 86% of them, meaning that the closest 9 matches presented by the model for each unknown included what we considered was the correct match. When chalky or degraded otoliths were excluded, the model returned 92% correct matches (Baker et al. in prep).

With further funding we will expand the library to more species and regions, and broaden the functionality, including allowing users to input the mass of their otolith to estimate the size of the fish it came from, based on otolith weight-fish length regressions we developed. Additional functions will allow users to upload and compare multiple images of otoliths to quantify symmetry.

Users may find the images and descriptions in Fossil Fishes of Alabama, Chapter 7, Otoliths (Ebersol et al. 2024) helpful for further identifications and for distinguishing closely related species.

Please cite as:
Baker R, David P, Heflin E, Lechervy A, Tschumperlé D (2026) Otolith Image Processing Library. www.disl.edu/research/oipl/

Questions, comments, feedback, want to contribute? Email: rbaker@disl.org

Attribution

This project was paid for [in part] with federal funding for the Alabama Center of Excellence from the Department of the Treasury under the Resources and Ecosystems Sustainability, Tourist Opportunities, and Revived Economies of the Gulf Coast States Act of 2012 (RESTORE Act) in cooperation with the State of Alabama Department of Conservation and Natural Resources under the Alabama Center of Excellence Program at the MESC/Dauphin Island Sea Lab.