Unsupervised Salient Object Extraction Based on Sparse Representation
Zhongsheng Li · 2012
The existing salient object extraction algorithms had high time complexity and didn't take the integrity of the objects into account,and a salient object extraction algorithm is proposed,which is adaptive to the resource-constrained environment.The mathematic model of sparse representation is built.The corresponding relation between the salient objects and spare representation,edge energy approximation pattern among regions,and gradual change pattern between adjacent regions are induced.The candidate regions are determined based on the relation,and the salient objects are locally extracted based on gradual change pattern and edge energy approximation pattern.The contrast experiments indicate that the salient objects are captured accurately and the integrity of the salient objects are well-kept under given conditions with the proposed algorithm.