Designing efficient alpha-tree algorithms
This PhD thesis explores how to make a widely used image‑analysis method, the alpha-tree, much faster and more efficient. Although alpha-trees are powerful tools for understanding complex images such as satellite photos or medical scans, they normally require a great deal of computation and memory. This makes them slow to use in practice, especially for large or high‑quality images.
The research introduces several new ways to speed up alpha-tree construction while reducing unnecessary computation. First, it presents a more efficient way of building the tree that avoids redundant steps and uses memory more efficiently. Tests on more than a thousand images show that this approach can more than double the speed while cutting memory waste dramatically.
The thesis also tackles situations where images contain extremely large differences in brightness or color, which normally slow down existing methods. A new strategy for organizing and processing image information makes the algorithm much faster in these challenging cases, improving performance by factors ranging from 1.7 to over 4 times depending on the image type.Finally, the work demonstrates how alpha-tree construction can be accelerated even further by running the process in parallel on modern multi‑core computers. Experiments on real satellite data show that the new parallel method can process tens of millions of pixels per second and achieve speedups of up to thirty times compared with traditional approaches.Together, these contributions make alpha-tree analysis significantly faster, more scalable, and more practical for real‑world applications.