Skip to content Search
Search our website:

A Self-tuning Approximation Clustering Algorithm and its Thresholding Versions for Delimitation of Upwelling on Sea Surface Temperature Images

  • Speaker: Dr. Susana Nascimento, Dep. of Computer Science and NOVA Laboratory for Computer Science and Informatics (NOVA LINCS), Faculdade de Ciências e Tecnologia, Universidade Nova de Lisboa, PORTUGAL
  • Date: Wednesday, 9 December 2015 from 16:00 to 17:00
  • Location: Room 151

In this talk, an algorithm derived from Mirkin’s Approximation Clustering model (MAC) is presented as a version of Seeded Region Growing (SRG) and applied for the automatic delimitation of coastal upwelling from Sea Surface Temperature (SST) images. The MAC algorithm, stands on a homogeneity criterion in the format of a product rather than the conventional difference between a pixel value and the mean of values over the region of interest. We introduce a self-tuning version of the algorithm in which the homogeneity threshold is locally derived from the approximation criterion over a window in the neighbourhood of pixels under consideration. The growing strategy considers a pixel-by-pixel boundary oriented window that serves as a regularizer for cluster growing.

We experimentally compared the self-tuning MAC algorithm with its versions embedding the popular image thresholding methods by Ridler, Kitller, and Otsu’s and applied these thresholding approximation clustering versions to automatically delineate coastal upwelling on SST images over different situations.

A comparative evaluation of the methods against manually labelled ground-truth maps by Oceanographers shows the superiority of the self-tuning MAC in case of strong gradient and noisy SST’s while the Kittler’s thresholding version of MAC wins in case of weak gradient SST’s.


[1] S. Nascimento, S. Casca, B. Mirkin (2015). A Seed Expanding Cluster Algorithm for Deriving Upwelling Areas on Sea Surface Temperature Images. Computers & Geociences, Special issue: Statistical learning in geoscience

This is Joint work with: master student Vasco Lopes (NOVA LINCS) and Boris Mirkin (Birkbeck University of London, UK)