Abstract
In this paper, we address the problem of combining multiple overlapping image sections of biological specimens to obtain a single image containing the entire specimen. This is useful in the digitisation of a large number of biological specimens stored in museum collections and laboratories. In the case of many large specimens, it means that the specimen must be captured in overlapping sections instead of a single image. In this research, we have compared the performance of several known algorithms for this problem. In addition, we have developed several new algorithms based on matching the geometry (width, slope, and curvature) of the specimens at the boundaries. Finally, we compare the performance of a bagging approach that combines the results from multiple stitching algorithms. Our detailed evaluation shows that brightness-based and curvature-based approaches produce the best matches for the images in this domain.
Original language | English (US) |
---|---|
Pages (from-to) | 1-17 |
Number of pages | 17 |
Journal | International Journal of Computational Vision and Robotics |
Volume | 2 |
Issue number | 1 |
DOIs | |
State | Published - 2011 |
Keywords
- biological specimen
- geometry-based matching
- image stitching
ASJC Scopus subject areas
- Computer Vision and Pattern Recognition
- Computer Science Applications