Research On Salient Object Dection In Complicted Environment

Salient object detection has been widely used in image processing, so the salient object detection is becoming a hot issue in recent years. The salient region in an image is an object which is familiar to people and gets the visual attention. The computer imitates the visual attention selection mechanism of human, to extract the most notable part and show as a gray image, to accomplish the salient object detection.Although the scholars at home and abroad have done a lot of research on salient detection, the precision has not reached people’s expection. Especially for the salient detection of a small object, two objects or the object detection under complex background is difficult. In this paper, we research the accuracy and correctness of the salient object detection in various complex environments. Based on summarizing the existing main detection algorithms, two algorithms are proposed for salient object detection in this paper.Firstly, salient object detection based on optimally fusing low-level features(FF) is proposed. Although the low-level features can detect salient object, it’s very difficult to deal with the image in a complicated environment. FF algorithm is optimally combining the low-level image features, making each feature complementary, so that it realizes salient object detection in the complex environment more accurately. This method not only effectively combines the global and local features, but also introduces the frequency information. FF effectively enhance the salient region and the boundary, solving the problem of the salient object uneven distributionSecondly, salient object detection based on background optimization(BO) is proposed. For the object detection, background information is also very important. BO algorithm is improved based on the image boundary prior to estimate the background, and then fuses the background and foreground information. The fusion is seemed as a global optimization function, and the optimal threshold is set according to the experiment. The optimal solution of the cost function is the saliency of the image. It ensures the detection accuracy and robustness, and the real time of detection has also been improved.Finally, in this paper, the four public data sets are used, respectively represent the images in a complicated environment(objects sizes are different, complex background, two objects, etc.). In addition, the two proposed algorithm conduce a comprehensive inspection in a large number of data sets, and compare with a variety of classic algorithms in the quantitative and qualitative way. The experimental results show that the FF and BO algorithm can realize salient object detection in the complex environment, and both have higher accuracy and correctness. In particular, BO algorithm has also been improved in real-time.