非监督层次化模糊相关的人体红外图像分割
|
摘要
针对当前多级模糊熵算法在分割人体红外图像时,存在划分数需人工指定,全局划分导致熵的信息度量精度受背景干扰,分割精度不高等问题,提出了非监督层次化模糊相关分割。首先采用熵率法将图像划分为若干超像素,确保区域一致性,提高后续处理效率;随后,用准确度量划分适当性的模糊相关来描述图像,构建模糊相关图割2-划分算子,提高层次化分割中单步分割的精度。2-划分算子的核心思想是利用提出的递推计算策略,快速搜索最大模糊相关时目标和背景的划分概率,并用其来设置图割的数据项,实施超像素的模糊相关图割2-划分。最后将2-划分算子与自顶向下的非监督层次化分割策略相结合,迭代地对目标超像素区域实施2-划分,自适应确定划分数,获得人体目标。实验结果表明:较常用算法,该算法不但能自动确定划分数,而且分割精度还提高了约18%,运行时间约为3.8s,能有效用于人体红外图像分割的工程实践中。
The multilevel fuzzy entropy has been adopted as a segmentation method for the human target infrared image.However,it has some problems that the partition number needs to be designated manually,and the global partition manner result in the accuracy of entropy measurement being influenced by background interference and the precision of segmentation being reduced.To address these problems,an unsupervised hierarchical segmentation based on fuzzy correlation strategy was proposed.To ensure the region homogeneity and improve efficiency,the entropy rate method was adoped to segment image into a group of superpixels.Then,the 2-partition segmentation operator through fuzzy correlation was built for improving precision of single step,since the fuzzy correlation can measure the appropriate partition well.The core idea of 2-partition segmentation operator was to utilize an iterative scheme to improve computational efficiency in fuzzy correlation evaluation.Then the probabilities of fuzzy events obtained by maximizing fuzzy correlation were used to define the data terms of graph cut.Finally,the 2-partition segmentation operator was combined with top-down hierarchical segmentation structure.By segmenting the superpixels of object regions by 2-partition segmentation operator iteratively,the partition number could be assigned and human target was achieved in this adaptive approach.The experiment results demonstrate that the proposed method can not only decide the partition number automatically,but can also improve the accuracy by 18%comparing with that of the existing methods and the running time is about 3.8 s.It can be used in engineering practice of human target infrared image segmentation.
The multilevel fuzzy entropy has been adopted as a segmentation method for the human target infrared image.However,it has some problems that the partition number needs to be designated manually,and the global partition manner result in the accuracy of entropy measurement being influenced by background interference and the precision of segmentation being reduced.To address these problems,an unsupervised hierarchical segmentation based on fuzzy correlation strategy was proposed.To ensure the region homogeneity and improve efficiency,the entropy rate method was adoped to segment image into a group of superpixels.Then,the 2-partition segmentation operator through fuzzy correlation was built for improving precision of single step,since the fuzzy correlation can measure the appropriate partition well.The core idea of 2-partition segmentation operator was to utilize an iterative scheme to improve computational efficiency in fuzzy correlation evaluation.Then the probabilities of fuzzy events obtained by maximizing fuzzy correlation were used to define the data terms of graph cut.Finally,the 2-partition segmentation operator was combined with top-down hierarchical segmentation structure.By segmenting the superpixels of object regions by 2-partition segmentation operator iteratively,the partition number could be assigned and human target was achieved in this adaptive approach.The experiment results demonstrate that the proposed method can not only decide the partition number automatically,but can also improve the accuracy by 18%comparing with that of the existing methods and the running time is about 3.8 s.It can be used in engineering practice of human target infrared image segmentation.
引文
[1]赵云峰,李夜金,张寅,等.海洋背景下运动目标的天基红外探测场景生成系统[J].光学精密工程,2017,25(4):1019-1025.ZHAO Y F,LI Y J,ZHANG Y,et al..A spacebased infrared detection scene generation system for moving objects with sea background[J].Opt.Precision Eng.,2017,25(4):1019-1025.(in Chinese)
[2]杨会玲,吴玉宏,孙慧婷,等.基于杂波抑制的海平线红外弱小目标检测[J].液晶与显示,2017,32(4):317-324.YANG H L,WU Y H,SUN H T,et al..Small dim infrared target detection of horizon region based on clutter rejection[J].Liquid Crystals and Displays,2017,32(4):317-324.(in Chinese)
[3]NIE F,ZHANG P,LI J,et al..An Image Segmentation Method Based on Renyi Relative Entropy and Gaussian Distribution[J].Recent Patents on Computer Science,2017,10(2):122-130.
[4]DENG H,SUN X,LIU M,et al..Entropy-based window selection for detecting dim and small infrared targets[J].Pattern Recognition,2017,61(2017):66-77.
[5]BENABDELKADER S,BOULEMDEN M.Recursive algorithm based on fuzzy 2-partition entropy for 2-level image thresholding[J].Pattern Recognition,2005,38(8):1289-1294.
[6]TANG Y G,MU W,ZHANG Y,et al..A fast recursive algorithm based on fuzzy 2-partition entropy approach for threshold selection[J].Neurocomputing,2011,74(17):3072-3078.
[7]ZHAN Y,PAN H,XIE X,et al..Graph entropybased clustering algorithm in medical brain image database[J].Journal of Intelligent&Fuzzy Systems,2016,31(2):1029-1039.
[8]尹诗白,王一斌,邓箴.基于快速递推模糊2-划分熵图割的红外图像分割[J].光学精密工程,2016,24(3):668-680.YIN S B,WANG Y B,DENG Z.Infrared image segmentation based on graph cut of fast recursive fuzzy 2-partition entropy[J].Opt.Precision Eng.,2016,24(3):668-680.(in Chinese)
[9]YIN S B,ZHAO X M,WANG W X,et al..Efficient multilevel image segmentation through fuzzy entropy maximization and graph cut optimization[J].Pattern Recognition,2014,47(9):2894-2907.
[10]YUAN J,BAE E,TAI X C.A study on continuous max-flow and min-cut approaches.Proceedings of the 27th Conference on the Computer Vision and Pattern Recognition,California,USA:CVPR,2010:2217-2224.
[11]BARISIONE F,SOLARNA D,DE GIORGI A,et al..Supervised classification of thermal infrared hyperspectral images through Bayesian,Markovian,and region-based approaches.Proceedings of the 36th Conference on Geoscience and Remote Sensing Symposium,Beijing,P.R.China:IGARSS,2016:937-940.
[12]TANG Y G,MU W W,ZHAO L X,et al..An image segmentation method based on maximizing fuzzy correlation and its fast recursive algorithm[J].Computers&Electrical Engineering,2014,40(3):833-843.
[13]LIU J and YANG Y H,Multiresolution color image segmentation[J].IEEE transactions on pattern analysis and machine intelligence,1994,16(7):689-700.
[14]LIU M Y,TUZEL O,RAMALINGAM S,et al..Entropy rate superpixel segmentation.Proceedings of 28th Conference on Computer Vision and Pattern Recognition,Colorado,USA:CVPR,2011:2097-2104.
[15]BOYKOV Y and KOLMOGOROV V.An experimental comparison of min-cut/max-flow algorithms for energy minimization in vision[J].IEEE transactions on pattern analysis and machine intelligence,2004,26(9):1124-1137.
[16]AJANTHAN T,HARTLEY R,SALZMANN M.Memory Efficient Max Flow for Multi-label Submodular MRFs.Proceedings of the 33th Conference on Computer Vision and Pattern Recognition,Las Vegas,USA:CVPR,2016:5867-5876.
[2]杨会玲,吴玉宏,孙慧婷,等.基于杂波抑制的海平线红外弱小目标检测[J].液晶与显示,2017,32(4):317-324.YANG H L,WU Y H,SUN H T,et al..Small dim infrared target detection of horizon region based on clutter rejection[J].Liquid Crystals and Displays,2017,32(4):317-324.(in Chinese)
[3]NIE F,ZHANG P,LI J,et al..An Image Segmentation Method Based on Renyi Relative Entropy and Gaussian Distribution[J].Recent Patents on Computer Science,2017,10(2):122-130.
[4]DENG H,SUN X,LIU M,et al..Entropy-based window selection for detecting dim and small infrared targets[J].Pattern Recognition,2017,61(2017):66-77.
[5]BENABDELKADER S,BOULEMDEN M.Recursive algorithm based on fuzzy 2-partition entropy for 2-level image thresholding[J].Pattern Recognition,2005,38(8):1289-1294.
[6]TANG Y G,MU W,ZHANG Y,et al..A fast recursive algorithm based on fuzzy 2-partition entropy approach for threshold selection[J].Neurocomputing,2011,74(17):3072-3078.
[7]ZHAN Y,PAN H,XIE X,et al..Graph entropybased clustering algorithm in medical brain image database[J].Journal of Intelligent&Fuzzy Systems,2016,31(2):1029-1039.
[8]尹诗白,王一斌,邓箴.基于快速递推模糊2-划分熵图割的红外图像分割[J].光学精密工程,2016,24(3):668-680.YIN S B,WANG Y B,DENG Z.Infrared image segmentation based on graph cut of fast recursive fuzzy 2-partition entropy[J].Opt.Precision Eng.,2016,24(3):668-680.(in Chinese)
[9]YIN S B,ZHAO X M,WANG W X,et al..Efficient multilevel image segmentation through fuzzy entropy maximization and graph cut optimization[J].Pattern Recognition,2014,47(9):2894-2907.
[10]YUAN J,BAE E,TAI X C.A study on continuous max-flow and min-cut approaches.Proceedings of the 27th Conference on the Computer Vision and Pattern Recognition,California,USA:CVPR,2010:2217-2224.
[11]BARISIONE F,SOLARNA D,DE GIORGI A,et al..Supervised classification of thermal infrared hyperspectral images through Bayesian,Markovian,and region-based approaches.Proceedings of the 36th Conference on Geoscience and Remote Sensing Symposium,Beijing,P.R.China:IGARSS,2016:937-940.
[12]TANG Y G,MU W W,ZHAO L X,et al..An image segmentation method based on maximizing fuzzy correlation and its fast recursive algorithm[J].Computers&Electrical Engineering,2014,40(3):833-843.
[13]LIU J and YANG Y H,Multiresolution color image segmentation[J].IEEE transactions on pattern analysis and machine intelligence,1994,16(7):689-700.
[14]LIU M Y,TUZEL O,RAMALINGAM S,et al..Entropy rate superpixel segmentation.Proceedings of 28th Conference on Computer Vision and Pattern Recognition,Colorado,USA:CVPR,2011:2097-2104.
[15]BOYKOV Y and KOLMOGOROV V.An experimental comparison of min-cut/max-flow algorithms for energy minimization in vision[J].IEEE transactions on pattern analysis and machine intelligence,2004,26(9):1124-1137.
[16]AJANTHAN T,HARTLEY R,SALZMANN M.Memory Efficient Max Flow for Multi-label Submodular MRFs.Proceedings of the 33th Conference on Computer Vision and Pattern Recognition,Las Vegas,USA:CVPR,2016:5867-5876.