Em-like soft segmentation for roman settlements detection in Switzerland

Castiello, Maria Elena; Cere, Raphael (28 August 2020). Em-like soft segmentation for roman settlements detection in Switzerland. In: 2020 Virtual Annual Meeting. European Association of Archaeologists

Text castiello abst.pdf - Published Version Restricted to registered users only Available under License Publisher holds Copyright. Download (1MB)

Spatial detection of object is a central issue in many disciplines as well as a major challenge in Archaeology. Such settlements detection can be assisted by Computer Vision approaches which offer a large body of state-of-the-art research notably in content-based image retrieval using graph-based segmentation.
In the large family of segmentation methods, the main approach considers objects as resulting from the aggregation of similar or homogeneous regions (nodes), according to their features dissimilarities (weighted edges) and their spatial relation.
Nowadays few formal researches in archaeology seem to have exploited the aforementioned clustering framework for archaeological site detection (features similarities and spatial proximity).
The present study is an interdisciplinary research project that combines archaeological knowledge with geographical and statistical expertise. We deal with roman ascertained archaeological evidences in Switzerland characterized by their spatial location and essential descriptive information. We used those characteristics to clarify whether an unexplored region of the study area may or may not contain the remains of ancient civilizations. We embedded the evidences in a regular grid and generated a new collection of regions. Each region is related to a numerical or categorical variable (e.g. environmental features) and holding a relative importance (i.e. weights). In order to reduce the redundancy of information in the data used, only the factorial scores as Principal Component Analysis results, were retained as features. Regions with “Presence” or “Absence” label are marked nodes respectively in distinct two groups, otherwise assigned to a third group (“Unknown”).
We have developed a EM-like soft segmentation iterative algorithm to maximize the model likelihood and to avoid separating strongly contiguous regions. This approach aims to attain a soft segmentation e.g. a regional clustering. The visual output is a Suitability map showing the probabilities that each region belongs to the group Presence – Absence – Unknown of archaeological evidences.

Interest & Impact

Downloads

1 since deposited on 16 Sep 2020
0 in the past 12 months

Citations

5 Citations in Web of Science ®
6 Citations in Scopus

Search

in Google Scholar™

Services

Actions (login required)

Edit item

Actions (login required)

Edit item