A Salient Object Detection Model Based on Local-region Contrast for Night Security and Assurance
Xia Wu, Yaling Wang, Zheming Zhang · 2019
Nowadays, night security and assurance are necessary and monitoring systems have been set up all over China. However, it is difficult to analyze the monitoring area because of the dark night light. In order to further maintain peace and ensure the safety and security of our study focused on the design of a more effective and robust approach to identify the salient area in a nighttime image. Therefore, this paper proposed an optimized salient object detection (SOD) model for night security and assurance. Firstly, enhance the image, then simple linear iterative clustering optimization (SLICO) method was used to divide the image into superpixel regions. Constructed the covariance matrix (CM) of the image based on various features, local saliency was calculated by non-linear fusion of these salient information. After this, integrated local saliency with the saliency of color space (CS) contrast. Finally, the regularization of the image was utilized to refine the salient graph. Experiments show that our method outperforms state of the art in the popular evaluation metrics.