Saliency Fusion in Eigenvector Space with Multi-Channel Pulse Coupled Neural Network

摘要

Saliency computation has become a popular research field for manyapplications due to the useful information provided by saliency maps. For asaliency map, local relations around the salient regions in multi-channelperspective should be taken into consideration by aiming uniformity on theregion of interest as an internal approach. And, irrelevant salient regionshave to be avoided as much as possible. Most of the works achieve thesecriteria with external processing modules; however, these can be accomplishedduring the conspicuity map fusion process. Therefore, in this paper, a newmodel is proposed for saliency/conspicuity map fusion with two concepts: a)input image transformation relying on the principal component analysis (PCA),and b) saliency conspicuity map fusion with multi-channel pulsed coupled neuralnetwork (m-PCNN). Experimental results, which are evaluated by precision,recall, F-measure, and area under curve (AUC), support the reliability of theproposed method by enhancing the saliency computation.