遥感影像主特征线检测
|
摘要
受到成像规律性差,背景纹理复杂及强烈噪声的影响,直线检测方法通常难以适应遥感影像处理的需求。有鉴于此,论文提出一种具有视觉显著性的遥感影像主特征线检测方法。论文首先论证了利用已提取直线为基元,基于格式塔法则构建主特征线的可行性;其次对直线在主特征线上的复杂投影情况进行了详细的剖析,并给出了主特征线的定义;接着建立了主特征线累计权重矩阵及直线统计矩阵,依据格式塔原则分析直线权重分布规律,以此构建了直线的权重模型,同时探讨不同直线在同一主特征线上权重分配规律;最后依据上述分析结果提出了具体的算法步骤。通过多幅含有强烈噪声的光学与SAR遥感卫星影像实验结果表明,相对于其他聚类算法,论文算法能够在杂乱无序的直线集中提取较为清晰的主特征线,并且实验效果基本符合人工视觉感知,便于机器对遥感影像的清晰理解。
The linear detection method is usually difficult to adapt to the demand of remote sensing image processing because of it exhibits poor imaging regularity, complex background texture, and strong noises. To address these problems, we proposed a new method that possesses visual saliency in the remote sensing image and can detect principal features. First, the possibility of constructing principal lines according to the Gestalt laws and the use of extracted lines as geometric primitives were analyzed. The complex projection of the lines in the principal lines was then examined, and the definition of the principal lines was provided. Furthermore, the cumulative weight matrix of the principal line and linear statistical matrix were constructed. Meanwhile, the distribution regularity of the short line weight was studied according to the Gestalt laws, and the model of the short linear weight was constructed. Accordingly, the weight allocation pattern of the different lines in the same principal line was also discussed. Finally, detailed algorithm steps were proposed according to these analyses. The key algorithm steps were described as follows: first, the chain code marshaling algorithms were employed to extract the straight lines. Second, the accumulative weighted matrix and linear statistical matrix of the principal lines were constructed. Third, the lines were sorted on the basis of their spatial positions. Fourth, according to linear weight distribution regularity, all the lines were elected to a cumulative weight matrix according to the linear weight model and distribution rules, and the results were recorded in the linear statistical matrix.Fifth, the local maximum value of the accumulative matrix was obtained to prevent parallel overlapping among the principal lines. Sixth,constraint analysis on the continuity and purity of the accumulative weighted matrix and linear statistical matrix was conducted to prevent the appearance of false principal lines. Finally, the parameters of the principal lines were obtained according to the sorting results of the weight voting matrix and weight values. Meanwhile, the principal line was obtained through its endpoints. The results of multiple SAR and optical remote sensing satellite images with strong noises showed that the traditional line extraction method can obtain only the disordered linear information, which is not clear and useful for image processing. In this study, our proposed method obtained clear principal lines by using Gestalt law on the basis of traditional linear extraction algorithm, and the results were basically in agreement with artificial visual perception. Meanwhile, the results suggest that our algorithm is superior to the traditional cluster algorithm in terms of operation efficiency and experimental effects. The experimental results indicate the potential application of our method in various fields, such as road extraction, image matching, and object recognition. However, this method also presents several shortcomings.First, the extraction results of the principal lines rely heavily on previous results. In addition, whether the linear-weighted Gaussian model established in this study is in full compliance with the Gestalt law requires further investigation. Finally, several parameter settings are experience values acquired by a large number of experiments. Thus, we hope to achieve the adaptive processing of these parameters in our future research.
The linear detection method is usually difficult to adapt to the demand of remote sensing image processing because of it exhibits poor imaging regularity, complex background texture, and strong noises. To address these problems, we proposed a new method that possesses visual saliency in the remote sensing image and can detect principal features. First, the possibility of constructing principal lines according to the Gestalt laws and the use of extracted lines as geometric primitives were analyzed. The complex projection of the lines in the principal lines was then examined, and the definition of the principal lines was provided. Furthermore, the cumulative weight matrix of the principal line and linear statistical matrix were constructed. Meanwhile, the distribution regularity of the short line weight was studied according to the Gestalt laws, and the model of the short linear weight was constructed. Accordingly, the weight allocation pattern of the different lines in the same principal line was also discussed. Finally, detailed algorithm steps were proposed according to these analyses. The key algorithm steps were described as follows: first, the chain code marshaling algorithms were employed to extract the straight lines. Second, the accumulative weighted matrix and linear statistical matrix of the principal lines were constructed. Third, the lines were sorted on the basis of their spatial positions. Fourth, according to linear weight distribution regularity, all the lines were elected to a cumulative weight matrix according to the linear weight model and distribution rules, and the results were recorded in the linear statistical matrix.Fifth, the local maximum value of the accumulative matrix was obtained to prevent parallel overlapping among the principal lines. Sixth,constraint analysis on the continuity and purity of the accumulative weighted matrix and linear statistical matrix was conducted to prevent the appearance of false principal lines. Finally, the parameters of the principal lines were obtained according to the sorting results of the weight voting matrix and weight values. Meanwhile, the principal line was obtained through its endpoints. The results of multiple SAR and optical remote sensing satellite images with strong noises showed that the traditional line extraction method can obtain only the disordered linear information, which is not clear and useful for image processing. In this study, our proposed method obtained clear principal lines by using Gestalt law on the basis of traditional linear extraction algorithm, and the results were basically in agreement with artificial visual perception. Meanwhile, the results suggest that our algorithm is superior to the traditional cluster algorithm in terms of operation efficiency and experimental effects. The experimental results indicate the potential application of our method in various fields, such as road extraction, image matching, and object recognition. However, this method also presents several shortcomings.First, the extraction results of the principal lines rely heavily on previous results. In addition, whether the linear-weighted Gaussian model established in this study is in full compliance with the Gestalt law requires further investigation. Finally, several parameter settings are experience values acquired by a large number of experiments. Thus, we hope to achieve the adaptive processing of these parameters in our future research.
引文
Ballard D H.1981.Generalizing the Hough transform to detect arbit rary shapes.Pattern Recognition,13(2):111-122[DOI:10.10160031-3203(81)90009-1]
Barni M,Cappellin V,Paoli A and Mecocci A.1996.Unsupervised de tection of straight lines through possibilistic clustering//Proceed ings of the International Conference on Image Processing Lausanne,Switzerland:IEEE:963-966[DOI:10.1109/ICIP1996.561065]
Burns J B,Hanson A R and Riseman E M.1986.Extracting straigh lines.IEEE Transaction on Pattern Analysis and Machine Intelli gent,PAMI-8(4):425-455[DOI:10.1109/TPAMI.1986.4767808]
Chen R J,Liu L G and Dong G C.2010.Detection of principal lines in images.Journal of Image and Graphics,15(3):403-408(陈仁杰刘利刚,董光昌.2010.图像主特征直线的检测算法.中国图象图形学报,15(3):403-408)[DOI:10.11834/jig.20100308]
Chung K L,Chen T C and Yan W M.2004.New memory-and compu tation-efficient Hough transform for detecting lines.Pattern Re cognition,37(5):953-963[DOI:10.1016/j.patcog.2003.09.008]
Cuneyt A and Cihan T.2011.EDLines:a real-time line segment detect or with a false detection control.Pattern Recognition Letters32(13):1633-1642[DOI:10.1016/j.patrec.2011.06.001]
Fiete R D and Tantalo T.2001.Comparison of SNR image quality met rics for remote sensing systems.Optical Engineering,40(4)574-585[DOI:10.1117/1.1355251]
Freeman H.1970.Boundary encoding and processing//Picture Pro cessing and Psychopictorics.New York:Academic Press241-266
Li C H,Shi H,Dai P Y,Chen J,Du X F,Qu Y Y and Xie Y.2011.Su per-resolution image reconstruction based on gestalt theory.Journ al of Xiamen University(Natural Science),50(2):261-270(李翠华,施华,戴平阳,陈婧,杜晓凤,曲延云,谢怡.2011.引入格式塔理论的超分辨率图像重建技术.厦门大学学报(自然科学版)50(2):261-270)
Stahl J S and Wang S.2008.Globally optimal grouping for symmetric closed boundaries by combining boundary and region information IEEE Transactions on Pattern Analysis and Machine Intelligence30(3):395-411[DOI:10.1109/TPAMI.2007.1186]
Tsuda K,Minoh M and Ikeda K.1996.Extracting straight lines by se quential fuzzy clustering.Pattern Recognition Letters,17(6)643-649[DOI:10.1016/0167-8655(96)00029-3]
Von Gioi R G,Jakubowicz J,Morel J M and Randall G.2010.LSD:a fast line segment detector with a false detection control.IEEETransactions on Pattern Analysis and Machine Intelligent,32(4)722-732[DOI:10.1109/TPAMI.2008.300]
Wang J X,Song W D,Zhao L K and Wang W X.2014a.Application of improved freeman chain code in edge tracking and straight line extraction.Journal of Signal Processing,30(4):422-430(王竞雪宋伟东,赵丽科,王伟玺.2014a.改进的Freeman链码在边缘跟踪及直线提取中的应用研究.信号处理,30(4):422-430)[DOI10.3969/j.issn.1003-0530.2014.04.008]
Wang J X,Zhu Q,Wang W X and Zhao L K.2014b.Straight line ex traction algorithm by Hough transform combining edge grouping Journal of Remote Sensing,18(2):378-389(王竞雪,朱庆,王伟玺,赵丽科.2014b.结合边缘编组的Hough变换直线提取.遥感学报,18(2):378-389)[DOI:10.11834/jrs.20143127]
Wu B,Zhang Y S and Zhu Q.2012.Integrated Point and Edge Match ing on Poor Textural Images Constrained by self-adaptive triangu lations.ISPRS Journal of Photogrammetry and Remote Sensing68:40-55[DOI:10.1016/j.isprsjprs.2011.12.005]
Xu W,Tang Z M,Xu D and Wu G X.2015.Integrating multi-features fusion and gestalt principles for pavement crack detection.Journa of Computer-Aided Design and Computer Graphics,27(1)147-156(徐威,唐振民,徐丹,吴国星.2015.融合多特征与格式塔理论的路面裂缝检测.计算机辅助设计与图形学学报,27(1)147-156)
Xu Z Z,Shin B S and Klette R.2015.Accurate and robust line seg ment extraction using minimum entropy with Hough transform IEEE Transactions on Image Processing,24(3):813-822[DOI10.1109/TIP.2014.2387020]
Zeng J X and Wang Y.2015.Natural image segmentation method based on Gestalt rules.Journal of Image and Graphics,20(8)1026-1034(曾接贤,王玉.2015.结合格式塔完形规则的自然图像分割.中国图象图形学报,20(8):1026-1034)[DOI10.11834/jig.20150805]
Barni M,Cappellin V,Paoli A and Mecocci A.1996.Unsupervised de tection of straight lines through possibilistic clustering//Proceed ings of the International Conference on Image Processing Lausanne,Switzerland:IEEE:963-966[DOI:10.1109/ICIP1996.561065]
Burns J B,Hanson A R and Riseman E M.1986.Extracting straigh lines.IEEE Transaction on Pattern Analysis and Machine Intelli gent,PAMI-8(4):425-455[DOI:10.1109/TPAMI.1986.4767808]
Chen R J,Liu L G and Dong G C.2010.Detection of principal lines in images.Journal of Image and Graphics,15(3):403-408(陈仁杰刘利刚,董光昌.2010.图像主特征直线的检测算法.中国图象图形学报,15(3):403-408)[DOI:10.11834/jig.20100308]
Chung K L,Chen T C and Yan W M.2004.New memory-and compu tation-efficient Hough transform for detecting lines.Pattern Re cognition,37(5):953-963[DOI:10.1016/j.patcog.2003.09.008]
Cuneyt A and Cihan T.2011.EDLines:a real-time line segment detect or with a false detection control.Pattern Recognition Letters32(13):1633-1642[DOI:10.1016/j.patrec.2011.06.001]
Fiete R D and Tantalo T.2001.Comparison of SNR image quality met rics for remote sensing systems.Optical Engineering,40(4)574-585[DOI:10.1117/1.1355251]
Freeman H.1970.Boundary encoding and processing//Picture Pro cessing and Psychopictorics.New York:Academic Press241-266
Li C H,Shi H,Dai P Y,Chen J,Du X F,Qu Y Y and Xie Y.2011.Su per-resolution image reconstruction based on gestalt theory.Journ al of Xiamen University(Natural Science),50(2):261-270(李翠华,施华,戴平阳,陈婧,杜晓凤,曲延云,谢怡.2011.引入格式塔理论的超分辨率图像重建技术.厦门大学学报(自然科学版)50(2):261-270)
Stahl J S and Wang S.2008.Globally optimal grouping for symmetric closed boundaries by combining boundary and region information IEEE Transactions on Pattern Analysis and Machine Intelligence30(3):395-411[DOI:10.1109/TPAMI.2007.1186]
Tsuda K,Minoh M and Ikeda K.1996.Extracting straight lines by se quential fuzzy clustering.Pattern Recognition Letters,17(6)643-649[DOI:10.1016/0167-8655(96)00029-3]
Von Gioi R G,Jakubowicz J,Morel J M and Randall G.2010.LSD:a fast line segment detector with a false detection control.IEEETransactions on Pattern Analysis and Machine Intelligent,32(4)722-732[DOI:10.1109/TPAMI.2008.300]
Wang J X,Song W D,Zhao L K and Wang W X.2014a.Application of improved freeman chain code in edge tracking and straight line extraction.Journal of Signal Processing,30(4):422-430(王竞雪宋伟东,赵丽科,王伟玺.2014a.改进的Freeman链码在边缘跟踪及直线提取中的应用研究.信号处理,30(4):422-430)[DOI10.3969/j.issn.1003-0530.2014.04.008]
Wang J X,Zhu Q,Wang W X and Zhao L K.2014b.Straight line ex traction algorithm by Hough transform combining edge grouping Journal of Remote Sensing,18(2):378-389(王竞雪,朱庆,王伟玺,赵丽科.2014b.结合边缘编组的Hough变换直线提取.遥感学报,18(2):378-389)[DOI:10.11834/jrs.20143127]
Wu B,Zhang Y S and Zhu Q.2012.Integrated Point and Edge Match ing on Poor Textural Images Constrained by self-adaptive triangu lations.ISPRS Journal of Photogrammetry and Remote Sensing68:40-55[DOI:10.1016/j.isprsjprs.2011.12.005]
Xu W,Tang Z M,Xu D and Wu G X.2015.Integrating multi-features fusion and gestalt principles for pavement crack detection.Journa of Computer-Aided Design and Computer Graphics,27(1)147-156(徐威,唐振民,徐丹,吴国星.2015.融合多特征与格式塔理论的路面裂缝检测.计算机辅助设计与图形学学报,27(1)147-156)
Xu Z Z,Shin B S and Klette R.2015.Accurate and robust line seg ment extraction using minimum entropy with Hough transform IEEE Transactions on Image Processing,24(3):813-822[DOI10.1109/TIP.2014.2387020]
Zeng J X and Wang Y.2015.Natural image segmentation method based on Gestalt rules.Journal of Image and Graphics,20(8)1026-1034(曾接贤,王玉.2015.结合格式塔完形规则的自然图像分割.中国图象图形学报,20(8):1026-1034)[DOI10.11834/jig.20150805]