作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2018, Vol. 44 ›› Issue (5): 240-245. doi: 10.19678/j.issn.1000-3428.0048490

• 图形图像处理 • 上一篇    下一篇

基于空间卷积神经网络模型的图像显著性检测

高东东,张新生   

  1. 西安建筑科技大学 管理学院,西安 710055
  • 收稿日期:2017-04-24 出版日期:2018-05-15 发布日期:2018-05-15
  • 作者简介:高东东(1990—),男,硕士研究生,主研方向为机器学习、图像显著性检测;张新生,教授。
  • 基金资助:
    陕西省自然科学基金(2016JM6023)。

Image Saliency Detection Based on Spatial Convolutional Neural Network Model

GAO Dongdong,ZHANG Xinsheng   

  1. School of Management,Xi’an University of Architecture and Technology,Xi’an 710055,China
  • Received:2017-04-24 Online:2018-05-15 Published:2018-05-15

摘要: 针对现有显著性检测方法鲁棒检测效果较差这一问题,提出一种新的基于空间卷积神经网络的显著性检测算法。利用去均值、归一化的预处理方法获取目标候选区。一方面通过引入卷积变换网络,建立提取显著物体上下文信息的全局模型,得到相应的目标检测信息显著图;另一方面构建特征子网络结构输出6维变换矩阵,经过空间变形模块改造输入图像,获取边缘信息。将空间变换网络输出的局部置信度融入到全局显著信息图,求取特征表达最大值,实现显著性与非显著性划分,完成显著性检测任务。实验结果表明,该算法不仅在同等条件下显著检测的AUC值得到了提高,并且生成的显著性图聚焦点突显,鲁棒检测效果得到明显改善。

关键词: 显著性检测, 特征融合, 卷积神经网络, 空间变换网络, 显著图

Abstract: To overcome the drawback of bad robustness detection of the existing saliency detection methods,this paper presents a novel saliency detection based on spatial convolutional neural network.The candidate object areas are conformed via preprocessing methods of the removing mean and normalization.By introducing the Convolutional Neural Network(CNN),a global model that captures the context information of saliency objects is constructed,and the corresponding target detection saliency map is obtained.On the other hand,the feature sub-network structure is established to output the 6-dimensional transformation matrix,which is used to transform input image through the spatial deformation module to obtain edge information of salient object.The output of the spatial transformer network local confidence coefficient is introduced into the global information saliency map to seek the maximum value of the feature expression.The discrimination of saliency and non-saliency is realized to accomplish the saliency detection task.Experimental results show that the proposed algorithm improves AUC accuracy under the same condition,generates a saliency map of focus highlighting and achieves impressive robust detection results.

Key words: saliency detection, feature fusing, Convolutional Neural Network(CNN), Spatial Transformer Network(STN), saliency map

中图分类号: