Advances in Connectivity and Connected Attribute Filters

In this paper we review recent advances in connected filtering, with emphasis on attribute filters based on component trees. We first describe basic connected filters using standard graph-based connectivity, based on the familiar 4, and 8-neighborhood relations in two dimensions. We show that connected filtering can be seen as a nonlinear counterpart of matching pursuit and other linear image representation schemes using overcomplete dictionaries. Understood this way, it can be shown that connected filters deliver a sparser image representation than do standard structural filters. We then show how abstractions of the notion of connectivity allow further manipulation of the dictionaries used, providing improved control over the perceptual groups detected in images. A review of algorithms to compute these filters is also presented. Beyond connectivity, we discuss hyperconnectivity and attribute-space connectivity, which offer ways to deal with overlapping structures efficiently while retaining much of the descriptive power of connectivity.

WILKINSON Michael;  OUZOUNIS Georgios; 

2010-09-28

Elsevier

JRC59391

9780123813183,   

1076-5670,   

http://www.sciencedirect.com/science/article/B7RNM-4YK6WTX-B/2/68da039cfccb4f488766957b35be2a07,    https://publications.jrc.ec.europa.eu/repository/handle/JRC59391,   

10.1016/S1076-567061005-1,