改进凸包的贝叶斯模型显著性检测算法

发布时间：2018-04-28 07:55

本文选题：显著性检测 + 凸包　；参考：《计算机辅助设计与图形学学报》2017年02期

【摘要】：针对目前显著性检测算法的准确性仍不是很理想的问题,提出改进已有的贝叶斯模型的显著性检测算法.首先利用图像压缩得到压缩图,结合经典的Harris算子来对原图和压缩图进行角点检测,利用角点得到两种图的最小凸包,将两者求交集来得到更合理的改进凸包;然后利用空间稀疏聚类算法结合改进凸包和超像素来得到先验图;再利用颜色直方图结合凸包来计算观察似然概率;最后根据已有的先验图和似然概率结合贝叶斯模型来得到显著性图,通过优化处理得到最终的显著性检测结果.在公开数据集MSRA和SED上进行仿真实验的结果表明,该算法不仅能够提高显著度图的视觉效果,而且查全率和查准率,F-measure,MAE等评价指标也比传统算法有明显提升.
[Abstract]:Aiming at the problem that the accuracy of the salience detection algorithm is still not ideal at present, a significance detection algorithm is proposed to improve the existing Bayesian model. Firstly, the compression graph is obtained by image compression, and the corner of the original graph and the compressed graph is detected by using the classical Harris operator. The minimum convex hull of the two graphs is obtained by using the corner point, and the intersection of the two graphs is obtained to obtain a more reasonable improved convex hull. Then, the priori graph is obtained by using spatial sparse clustering algorithm combined with improved convex hull and hyperpixel, and the observed likelihood probability is calculated by using color histogram and convex hull. Finally, according to the prior graph and likelihood probability combined with Bayesian model, the significance graph is obtained, and the final significance detection result is obtained by optimization processing. The simulation results on open datasets MSRA and SED show that the proposed algorithm can not only improve the visual effect of saliency map, but also improve the evaluation indexes such as recall rate and precision ratio and F-measurement mae.
【作者单位】：上海理工大学光电信息与计算机工程学院;上海师范大学信息与机电工程学院;上海交通大学电子信息与电气工程学院;
【基金】：国家自然科学基金联合项目(U1304616);国家自然科学基金青年项目(61502220);国家自然科学基金面上项目(61472245) 东方学者基金(15HJPY-MS02) 上海市自然科学基金(15ZR1428600)
【分类号】：TP391.41

【相似文献】