基于边缘引导的光场图像显著性检测  

Edge-guided light field image saliency detection

作　　者：梁晓 邓慧萍[1,2] 向森 吴谨[1,2] LIANG Xiao;DENG Hui-ping;XIANG Sen;WU Jin(School of Information Science and Engineering,Wuhan University of Science and Technology,Wuhan 430081,China;Engineering Research Center for Metallurgical Automation and Measurement Technology of Ministry of Education,Wuhan University of Science and Technology,Wuhan 430081,China)

机构地区：[1]武汉科技大学信息科学与工程学院,湖北武汉430081 [2]武汉科技大学、冶金自动化与检测技术教育部工程研究中心,湖北武汉430081

出　　处：《液晶与显示》2023年第5期644-655,共12页Chinese Journal of Liquid Crystals and Displays

基　　金：国家自然科学基金(No.61702384,No.61502357)。

摘　　要：针对光场图像显著性检测存在检测目标不完整、边缘模糊的问题,本文提出了一种基于边缘引导的光场图像显著性检测方法。利用边缘增强网络提取全聚焦图像的主体图和边缘增强图,结合主体图和焦堆栈图像所提取的特征获得初始显著图,以提高检测结果的准确性和完整性;将初始显著图和边缘增强图通过特征融合模块进一步学习边缘特性的信息,突出边缘细节信息;最后,使用边界混合损失函数优化以获得边界更为清晰的显著图。实验结果表明,本文所提出的网络在最新的光场图像数据集上,F-measure和MAE分别为0.88和0.046,表现均优于现有的RGB图像、RGB-D图像和光场图像显著性检测算法。所提方法能够更加精确地从复杂场景中检测出完整的显著对象,获得边缘清晰的显著图。Aiming at the problems of incomplete detection targets and blurred edges in light field image saliency detection,this paper proposes an edge-guided light field image saliency detection method.The edge enhancement network is used to extract the main image and edge enhancement image of all-focus image,and the initial saliency map is obtained by combining the features extracted from the main image and focal stack image to improve the accuracy and completeness of detection results.The initial saliency map and edge enhancement further learns the information of edge characteristics and highlights the edge details through the feature fusion module.Finally,the boundary mixing loss function is used to optimize the saliency map with clearer boundaries.The experimental results show that on the latest light field data set,F-measure and MAE are 0.88 and 0.046 respectively,which are better than the existing RGB images,RGB-D images and light field image saliency detection algorithms.The proposed method can more accurately detect complete salient objects from complex scenes,and obtain saliency maps with clear edges.

关 键 词：显著性检测 深度学习 光场图像 卷积神经网络 边缘检测网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]