Saliency Object Detection Algorithm Based On Proposals

With the rapid development of the Internet,image information grows explosively,which has become more and more demanding for effective image screening.The saliency object detection is to simulate the human visual attention mechanism to extract the most interesting area in the image,so that the computer can pay more attention to the foreground and ignore the background to improve the computational efficiency.Therefore,the research of saliency object detection has become more and more important in the field of image processing and computer vision in recent years.Firstly,in view of the shortcomings of which saliency object detection model results exist partial background region,the paper puts forward a kind of saliency object detection model that implements background pre-selection based on object proposals.The experimental results show that the model enhances the performance of the foreground object while expanding the background region,thereby effectively suppressing the background,highlight the prospect and improving the performance of the saliency detection.Secondly,in view of the above model is only on the basis of object proposals regions of the score,probably determine the position of the saliency object,did not use each object proposals regions information characteristics,a model for solving the saliency detection problem by multi-example learning is proposed.The object proposals are taken as the bags of multi-instance learning,and the instances in the bags are the super-pixels included in the object proposals.At the same time,an energy function is used to combine the appearance similarity,spatial structure information and global uniqueness in the image into an objective function to optimize the detection results.The experimental results show that the model greatly improves the saliency detection effect.Finally,in view of the problem that the absorbing Markov chain model accurately select saliency nodes is difficult,a salient object detection model based on the absorbed Markov chain with depth feature is proposed.The model use the image depth feature extracted from the depth network to get the transition probability matrix of absorbing Markov chain model.In addition,an angular embedding technique is applied to improve the saliency results,the experimental results show that the model has achieved good results on the public datasets.