Salient Object Detection Method And Application Based On Prior Integration

Saliency detection is a very important and challenging research area in computer vision.And it can be utilized into many computer vision areas,such as object detection,image compression,image segmentation and so on.In recent years,saliency detection has caused wide attention from a large number of domestic and foreign researchers.Referencing prior integration,salient object detection is studied deeply in this paper,and applied to the field of light background photo enhancement and traffic signs recognition.The main contains and innovation of this paper are as follows:(1)In order to obtain better result by using the complementary relationship between different prior,a novel salient object detection algorithm is proposed based on integration of center-bias prior and background-bias prior in chapter three.Firstly,a background-bias prior map is constructed via absorbing Markov chain with treating superpixels on the borders as the absorbing states and computing expected number of steps to obtain background-bias prior values.Secondly,a center-bias prior map is constructed based on a two-dimensional Gaussian function which peak is at the image center marked by a novel Harris corner detection algorithm.These two prior maps are combined as the final saliency map.At meantime,results from multi-scale saliency maps are integrated to further improve the detection performance.Extensive experiments against 10 state-of-art methods are carried out on ASD,SED1,SED2 and SOD benchmark datasets.Experimental results show that the proposed method performs favorably in terms of precision and recall.The visual comparison shows that prior integration can highlight the salient object in the image compared with a single prior information.(2)A light background photo enhancement algorithm based on visual saliency is proposed in chapter four.According to the Markov absorbing chain,the multi-scale saliency values of the image is calculated by different background node extraction method.The area to be enhanced is divided by Otsu method,and then it is enhanced in CIELab color model.The effectiveness of the proposed algorithm is verified by experiments on the light background photo.The experimental results show that the selection of background nodes based on half edge and gray level is reasonable.The saliency map can be used to accurately segment the area to be enhanced,so as to achieve the purpose of enhancing the light background photo.(3)A detection and recognition algorithm of traffic signs based on visual saliency is proposed in chapter five.The traffic sign is located by the saliency detection method,and segmented by Otsu method and morphological filter.To recognize traffic signs,color and shape detection are applied synthetically.In order to verify the effectiveness of the proposed algorithm,the experiment is carried out on the traffic sign database which is built by the project team.The experimental results show that the proposed algorithm has high accuracy and practical value.The experimental results show that the proposed algorithm has high accuracy.And the polygon detection algorithm based on edge-center distance can identify the shape quickly and accurately with robustness.