Saliency Detection Based On Markov Chain And Adaboost

With the development of science and technology,experts found that the human has a visual attention mechanism which can help people quickly extract effective information in the complicated scenes.A lot of researchers are trying to introduce the visual attention mechanism into computer vision,because it is useful to analysis images and reduce the human workload.In recent years,image saliency detection as one of the research directions in the field of image processing,has been successfully applied in many fields,such as object detection,object recognition,image compression and so on.Traditional salient object detection methods used to focus on low level features and prior knowledge.When the salient objects appear in picture edge,the results generated by the traditional methods are usually not ideal.In this paper,an effective saliency detection method based on Markov Chain and Adaboost is proposed to minimize the error caused by boundary prior.The main work includes the following aspects:(1)An initial saliency map is generated by absorbing Markov chain model at first.In this model,the transient probability is defined by the difference of color and texture between the super-pixels.Transient nodes ' absorbed time are calculated as their saliency value.The initial saliency map is likely to detect fine details.(2)A strong classifier optimization model based on Adaboost is proposed by us.The initial saliency map is optimized by the trained Adaboost model to get a optimized saliency map which focus on global contrast.By combining the two maps we get the final saliency map which combine the advantages of the above two maps.(3)We use method based on Conditional Random Fields to segment the salient region.This method is more flexible than the traditional threshold segmentation method.The complete and clear salient regions can be obtained by this method.The experimental results on four benchmark datasets(MSRA?ASD?SED?SOD)show that the proposed methods perform favorably against six modern saliency methods in tems of some popular evaluation measures such as the Precision and Recall curve(P-R curve),AUC(Area Under ROC Curve)and F-measure value.The experimental results show that optimization model which we proposed can be easily applied to other saliency models to improve their performance.A lot of experiments on benchmark datasets demonstrate the robustness and efficiency of the proposed method against other methods.