PROJECT TITLE :

Learning a combined model of visual Saliency for fixation prediction - 2016

ABSTRACT:

A giant variety of saliency models, every primarily based on a totally different hypothesis, are proposed over the past twenty years. In practice, whereas subscribing to one hypothesis or computational principle makes a model that performs well on some sorts of images, it hinders the final performance of a model on arbitrary images and large-scale data sets. One natural approach to enhance overall saliency detection accuracy would then be fusing different types of models. In this paper, galvanized by the success of late-fusion ways in semantic analysis and multi-modal biometrics, we propose to fuse the state-of-the-art saliency models at the score level in a very para-boosting learning fashion. 1st, saliency maps generated by many models are used as confidence scores. Then, these scores are fed into our para-boosting learner (i.e., support vector machine, adaptive boosting, or chance density estimator) to generate the final saliency map. So as to explore the strength of para-boosting learners, ancient transformation-primarily based fusion strategies, like Add, Min, and Max, are explored and compared during this paper. To more reduce the computation cost of fusing too many models, solely some of them are thought of in the next step. Experimental results show that score-level fusion outperforms every individual model and can any cut back the performance gap between the current models and the human inter-observer model.