A Bayesian Architecture for Combining Saliency Detectors

Saliency mechanisms can play an important role in the ability of recognition systems to deal with cluttered scenes. Saliency detection has also been an active area of computer vision, where existing solutions can be divided into two major classes: domain-independent and domain-dependent. In this work, it is proposed that the two classes are complementary and can be addressed within a unified formulation of the saliency problem, inspired on regularization theory. A Bayesian saliency framework is then proposed, in which domain-independent saliency maps are interpreted as priors for salient location, which can be used to regularize estimates of salient point location derived with domain-dependent procedures. Saliency maps are modeled as mixture distributions, and an analytical solution derived for the posterior distribution of true salient locations given the observed saliency measurements. This framework is shown to enable explicit control over the relative importance of the two types of saliency, reveals an interpretation of domain-dependent saliency as a focus-of-attention mechanism, and has a simple non-parametric extension. Experimental evaluation demonstrates the benefits of Bayesian saliency for weakly supervised recognition problems. Author email: [email protected] c ©University of California San Diego, 2005 This work may not be copied or reproduced in whole or in part for any commercial purpose. Permission to copy in whole or in part without payment of fee is granted for nonprofit educational and research purposes provided that all such whole or partial copies include the following: a notice that such copying is by permission of the Statistical Visual Computing Laboratory of the University of California, San Diego; an acknowledgment of the authors and individual contributors to the work; and all applicable portions of the copyright notice. Copying, reproducing, or republishing for any other purpose shall require a license with payment of fee to the University of California, San Diego. All rights reserved. SVCL Technical reports are available on the SVCL’s web page at http://www.svcl.ucsd.edu University of California, San Diego Statistical Visual Computing Laboratory 9500 Gilman Drive, Mail code 0407 EBU 1, Room 5512 La Jolla, CA 92093-0407