全局形状与局部一致性联合建模的显著性检测信息与通信工程专业论文.docxVIP

Saliency Saliency Detection via Jointly Modeling Global Shape and Local Consistency Abstract Humans have the ability tO quickly identify the salient object or region in complex image， and prepare for the further processing of the image of the human visual system．The task of visual saliency iS simulating human visual system by computer to locate salient object and region，and exporting saliency map indicating the probability of pixel belongs to salient object and region．Visual saliency detection is a fundamental problem in computer vision，which plays an important role in image compression，object recognition，image retrieval and SO on· In this paper。two saliency models are proposed based on Restricted Boltzmann Machine(RBM)and Conditional Random Field(CRF)．These two models detect salient object globally and locally，respectively．However，traditional global and local saliency detection algorithms have different detection drawbacks because of their different emphases．Based on this，we propose tWO strategies integrating the above-mentioned models and improving the detecting performance．Firstly，we pre—segment the given images mto 1mage segments different scales and extract appearance feature for image segments．Secondly，we build the global model based on RBM to get the global object shape prior，and the local model based on CRF to enforce the 10cal consistency．Finally，we proposed two integration mechanism to generate the final saliency maps．The first mechanism worked by jointly modeling CRF and RBM，to get the hybrid model，and testing on the testing lmage dataset，the second mechanism worked by solving cost function to integrate two saliency maps generated by the two above．mentioned models．The two integration mechanisms can retain the advantages ot global and local models，and generate better saliency maps． In the experiment，the proposed model is compared with twenty—three state—of-the—art saliency algorithms on five public saliency detection databases．Experimental results show