基于前馈上下文和形状先验的平面标注方法

doi:10.11896／j.issn.1002-137X.2018.12.039

计算机科学 ›› 2018, Vol. 45 ›› Issue (12): 235-242.doi: 10.11896／j.issn.1002-137X.2018.12.039

基于前馈上下文和形状先验的平面标注方法

郭燕飞¹, 刘宏哲¹, 袁家政^1,2, 王雪峤¹

(北京联合大学北京市信息服务工程重点实验室北京100101)¹
(北京开放大学北京100081)²

收稿日期:2017-11-10 出版日期:2018-12-15 发布日期:2019-02-25
作者简介:郭燕飞(1992-),女,硕士生,主要研究方向为数字图像处理,E-mail:2228133971@qq.com;刘宏哲(1971-),女,教授,主要研究方向为语义计算、数字博物馆、分布式系统集成;袁家政(1970-),男,教授,主要研究方向为图形图像处理、文物遗迹的数字化处理、数字博物馆等,E-mail:yuanjz@bjou.edu.cn(通信作者);王雪峤(1986-),女,讲师,主要研究方向为人脸识别。
基金资助:
本文受国家自然科学基金(61571045,61372148),北京市自然科学基金(4152016),国家科技支撑项目:“多彩贵州”文化资源集成与文化旅游综合服务应用示范(2015BAH55F03),北京市属高校高水平教师队伍建设创新团队建设提升计划(IDHT20170511)资助。

Surface Labeling Method Based on Feed-forward Context and Shape Priors

GUO Yan-fei¹, LIU Hong-zhe¹, YUAN Jia-zheng^1,2, WANG Xue-jiao¹

(Beijing Key Laboratory of Information Service Engineering,Beijing Union University,Beijing 100101,China)¹
(Beijing Open University,Beijing 100081,China)²

Received:2017-11-10 Online:2018-12-15 Published:2019-02-25

摘要/Abstract

摘要： 针对真实场景中由于互相遮挡导致的场景语义不能完全被理解的问题,提出了一种基于前馈上下文和形状先验的方法来对前景区域和被遮挡的背景区域进行语义标注。首先,将原始图像分割成超像素并提取像素点特征,采用加速决策树方法标注前景,同时采用改进的基于多尺度可形变的部件模型方法进行目标检测。其次,将可见对象信息与前馈上下文预测相结合来推测背景区域的被遮挡部分。然后,根据与当前标签置信度相匹配的多边形为每个标签提供形状先验知识。最后,结合像素预测与可视平面预测和多边形知识,以形成完整的场景标注图像。与现有方法相比,该方法能够得到与街道场景更相符的结果,并在人行道和公路较接近时的标注效果更好。

关键词: 场景理解, 多尺度可变的部件模型, 平面标注, 前馈上下文, 形状先验

Abstract: Aiming at the problem that the scene semantics cannot be understood caused by mutual occlusion,this paper proposed a method based on feed-forward context and shape priors to semantically label the foreground region and the occluded background area.Firstly,the original image is divided into super pixels,and the feature of pixel is extracted.The accelerated decision tree method is used to mark the foreground and the target model is detected by the improved multi-scale deformable component model method.Then,the visible object information is combined with the feed-forward context prediction to infer the occluded portions of background region.Next,the prior knowledge of shape for each label is provided based on polygons which match the current label confidence.Finally,the pixel prediction is combined with the visual plane prediction and the polygon knowledge to form a complete scene labeling image.Compared with the exis-ting method,this method can get more consistent results with the street scene,and can perform better labeling effect when the sidewalk is close to the road.

Key words: Feed-forward context, Multi-scale deformable component model, Scene understanding, Shape priors, Surface labeling

中图分类号:

TP391

郭燕飞, 刘宏哲, 袁家政, 王雪峤. 基于前馈上下文和形状先验的平面标注方法[J]. 计算机科学, 2018, 45(12): 235-242. https://doi.org/10.11896／j.issn.1002-137X.2018.12.039

GUO Yan-fei, LIU Hong-zhe, YUAN Jia-zheng, WANG Xue-jiao. Surface Labeling Method Based on Feed-forward Context and Shape Priors[J]. Computer Science, 2018, 45(12): 235-242. https://doi.org/10.11896／j.issn.1002-137X.2018.12.039

参考文献

[1]SOULY N,SHAH M.Scene labeling using sparse precision matrix[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2016:3650-3658.
[2]LADICKY′ L,STURGESS P,ALAHARI K,et al.What,where and how many? combining object detectors and crfs[C]∥European conference on computer vision.Springer Berlin Heidelberg,2010:424-437.
[3]LIU C,YUEN J,TORRALBA A.Nonparametric scene parsing:Label transfer via dense scene alignment[C]∥IEEE Conference on Computer Vision and Pattern Recognition.IEEE,2009:1972-1979.
[4]TIGHE J,LAZEBNIK S.Superparsing:scalable nonparametric image parsing with superpixels[J].Eruopean Conference on Computer Vision,2010,101(2):352-365.
[5]EIGEN D,FERGUS R.Nonparametric image parsing usingadaptive neighbor sets[J].Computer vision and pattern recognition,2012,157(10):2799-2806.
[6]SINGH G,KOSECKA J.Nonparametric scene parsing withadaptive feature relevance and semantic context[C]∥Procee-dings of the IEEE Conference on Computer Vision and Pattern Recognition.2013:3151-3157.
[7]LECUN Y,BOSER B,DENKER J S,et al.Backpropagation applied to handwritten zip code recognition[J].Neural Computation,1989,1(4):541-551.
[8]FARABET C,COUPRIE C,NAJMAN L,et al.Learning Hiera-rchical Features for Scene Labeling[J].IEEE Transactions on Pattern Analysis & Machine Intelligence,2013,35(8):1915-1929.
[9]LONG J,SHELHAMER E,DARRELL T.Fully convolutional networks for semantic segmentation[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2015:3431-3440.
[10]SHUAI B,WANG G,ZUO Z,et al.Integrating parametric and non-parametric models for scene labeling[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2015:4249-4258.
[11]PINHEIRO P,COLLOBERT R.Recurrent convolutional neural networks for scene labeling[C]∥International Conference on Machine Learning.2014:82-90.
[12]LIANG M,HU X,ZHANG B.Convolutional neural networks with intra-layer recurrent connections for scene labeling[C]∥Advances in Neural Information Processing Systems.2015:937-945.
[13]HOIEM D,EFROS A A,HEBERT M.Recovering Surface La-yout from an Image[J].International Journal of Computer Vision,2007,75(1):151-172.
[14]FZLZENSZWALB P F,GIRSHICK R B,MCALLESTER D,et al.Ojbect Detection with Discriminatively Trained Part-Based Models.IEEE Transactions on Pattern Analysis & Machine Intelligence,2010,32(9):1627-1645.
[15]GRIBBON K T,BAILEY D G.A novel approach to real-time bilinear interpolation[C]∥Second IEEE International Workshop on Electronic Design.IEEE,2004:126-131.
[16]HWANG J W,LEE H S.Adaptive image interpolation based on local gradient features[J].IEEE Signal Processing Letters,2004,11(3):359-362.
[17]TU Z.Auto-context and its application to high-level vision tasks[C]∥IEEE Conference on Computer Vision and Pattern Recognition,2008(CVPR 2008).IEEE,2008:1-8.
[18]SHOTTON J,JOHNSON M,CIPOLLA R.Semantic texton fore-sts for image categorization and segmentation[C]∥IEEE Conference on Computer Vision & Pattern Recognition.2008:1-8.
[19]ZHANG H,XIAO J,QUAN L.Supervised Label Transfer for Semantic Segmentation of Street Scenes∥European Confer-ence on Computer Vision.2010:561-574.
[20]BYEON W,BREUEL T M,RAUE F,et al.Scene labeling with LSTM recurrent neural networks[C]∥Computer Vision and Pattern Recognition.IEEE,2015:3547-3555.
[21]TIGHE J,LAZEBNIK S.Finding things:Image parsing with re-gions and per-exemplar detectors[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2013:3001-3008.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

基于前馈上下文和形状先验的平面标注方法

Surface Labeling Method Based on Feed-forward Context and Shape Priors

PDF (PC)

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 3

Metrics

本文评价

推荐阅读 0

[1]	姚拓中, 左文辉, 安鹏, 宋加涛. 基于多重语义交互的递归式场景理解框架 Multi-semantic Interaction Based Iterative Scene Understanding Framework 计算机科学, 2019, 46(5): 228-234. https://doi.org/10.11896/j.issn.1002-137X.2019.05.035
[2]	杨利萍，邹琪. 基于先验形状信息的水平集图像分割 Level Set Image Segmentation Method Based on Prior Shape Knowledge 计算机科学, 2012, 39(8): 288-291.
[3]	梁浩哲,李国辉,张军. 基于运动轨迹的监控场景分析模型 Surveillance Scene Analysis Model Based on Motion Trajectory 计算机科学, 2011, 38(9): 264-266.