基于背景评估的贝叶斯模型显著性检测

A Bayesian Model Saliency Detection Algorithm Based on Background Information Evaluation

针对自然图像中,复杂背景信息对显著性目标检测的影响,提出一种利用背景信息进行预测和贝叶斯模型选择优化的显著性检测方法。首先,为了提取完整的先验信息,根据背景信息与图像边界的连通性,以及对图像边界是否为背景进行评估来生成先验显著图。其次,为了降低背景信息的干扰,通过对流行排序算法生成的显著图进行角点检测,选择较为准确的显著点构造凸包。最后,利用贝叶斯模型进行选择优化来抑制和显著目标具有相同特征的背景信息。在2个公开的数据集上进行测试,并与4种性能较好的显著性检测算法对比,结果显示本文算法可提高显著性检测的准确性和区域的完整性。

Aiming at the influence of complex background information on salient object detection in natural images,this paper proposes a saliency detection method based on background information prediction and Bayesian model selection optimization.First,in order to extract complete prior information,a prior saliency map is generated according to the evaluation of the connectivity between the background information and the image boundary,and whether the image boundary is the background.Secondly,in order to reduce the interference of background information,corner detection is performed on saliency map generated by popular sorting algorithm,and the more accurate salieney points are selected to construct convex hull.Finally,Bayesian model is used for selection optimization to suppress the background information with the same characteristics as the salient object.Experiment is tested on two public datasets and compared with four classical saliency detection algorithms.The results show that the proposed algorithm can improve the accuracy of saliency detection and regional integrity.