張滿 顏普

摘要：該文提出了一種基于圖割和支持向量機(jī)(SVM)的彩色遙感圖像分割算法，首先針對在RGB空間中很難有效區(qū)分顏色相似性的問題,通過選擇更加符合顏色視覺特性的HSI顏色空間進(jìn)行圖像處理與分析，然后利用圖割算法定義一個(gè)能量函數(shù)來判斷圖像中像素是屬于圖像前景還是圖像背景，從而有選擇的提取前景種子節(jié)點(diǎn)與背景種子節(jié)點(diǎn)。最后以前景種子節(jié)點(diǎn)與背景種子節(jié)點(diǎn)為SVM算法的訓(xùn)練集對遙感圖像進(jìn)行訓(xùn)練分類，完成圖像的分割。實(shí)驗(yàn)表明此方法是一種有效的彩色遙感圖像分割方法。

關(guān)鍵詞：圖像分割；支持向量機(jī)；圖割圖論；HIS

中圖分類號：TP391文獻(xiàn)標(biāo)識碼：A文章編號：1009-3044(2012)16-3958-04

A Color Remote Sensing Image Segmentation Algorithm Based on Graph Cut and Support Vector Machine

ZHANG Man，YAN Pu

（Key Lab Intelligent Computing & Signal Ministry of Education,Anhui University, Hefei 230039,China)

Abstract：A color remote sensing image segmentation algorithm based on graph cut and support vector machine is proposed in this paper. Firstly, because it is difficulty to evaluate the similarity of two colors from their distance in RGB color space, color image processing is im? plemented in Hue Saturation Intensity (HSI) space which reflects the features of human vision. Secondly, An energy function is defined by graph cut algorithm for determining the image pixels belongs to the image background image and the prospects,then prospects seed nodes and background seed nodes are used as training set of SVM.Finally, remote sensing image are classified using SVM. A color remote sensing image segmentation is completed. Experiments show the efficiency of the proposed algorithm on color remote sensing images.

Key words：Image Segmentation; Support Vector Machine (SVM); Graphic Theory and Graph Cut; HIS

在對圖像的研究和應(yīng)用中，人們往往僅對圖像中的某些部分感興趣，這些部分常稱為目標(biāo)或前景（其他部分稱為背景），它們一般對應(yīng)圖像中特定的具有獨(dú)特性質(zhì)的區(qū)域。圖像分割就是把圖像空間劃分成若干個(gè)具有某些一致性屬性的不重疊區(qū)域并提取出感興趣目標(biāo)的技術(shù)和過程，而對圖像空間的劃分建立在區(qū)域的相似性和非連續(xù)性基礎(chǔ)上。相似性就是說同一區(qū)域中的像素特征是類似的；非連續(xù)性表明不同區(qū)域間像素的特征存在突變。

圖像分割是從圖像中分割出感興趣的區(qū)域。典型的圖像分割算法包括基于區(qū)域的算法[1-3]和基于輪廓的算法[4-6]，而基于圖論的圖像分割技術(shù)是近年來圖像分割領(lǐng)域的一個(gè)研究熱點(diǎn)[7-9]。遙感圖像具有數(shù)據(jù)量大、模糊性較強(qiáng)、紋理細(xì)節(jié)豐富等特點(diǎn)，這就決定了無論在分割效率還是分割效果上都對遙感圖像的分割提出了比自然景色圖像分割更高的要求。

該文提出了一種基于圖割和支持向量機(jī)(SVM)的彩色遙感圖像分割算法，首先選擇更加符合顏色視覺特性的HSI顏色空間進(jìn)行圖像處理與分析，然后利用圖割算法定義一個(gè)能量函數(shù)來判斷圖像中像素是屬于圖像前景還是圖像背景，從而有選擇的提取前景種子節(jié)點(diǎn)與背景種子節(jié)點(diǎn)。最后用前景種子節(jié)點(diǎn)和背景種子節(jié)點(diǎn)作為訓(xùn)練集，并對遙感圖像進(jìn)行SVM訓(xùn)練分類，之后進(jìn)行必要的后期處理，最終完成圖像的分割。

該文提出了一種基于圖割和支持向量機(jī)的彩色遙感圖像分割算法，利用圖割算法來判斷圖像中像素是屬于圖像前景還是圖像背景，從而有選擇的提取前景種子節(jié)點(diǎn)與背景種子節(jié)點(diǎn),以前景種子節(jié)點(diǎn)與背景種子節(jié)點(diǎn)為訓(xùn)練集對遙感圖像利用SVM進(jìn)行訓(xùn)練分類，最終完成圖像的分割，不僅消除圖割中產(chǎn)生的邊界模糊現(xiàn)象，且極大的提高了圖論方法分割的速度。實(shí)驗(yàn)證明了該文算法速度快精度準(zhǔn)，是一種有效的遙感圖像分割方法。

[1] Wang Zhongwu, Jensen J R, Im J. An automatic region-based image segmentation algorithm for remote sensing applications[J]. Environ? mental Modelling & Software, 2010,25(10) : 1149-1165.

[2] Jifeng Ning, Lei Zhang, Zhang D, et al. Interactive image segmentation by maximal similarity based region merging[J]. Pattern Recogni? tion, 2010,43(2): 445-456.

[3] Carvalho E A, Ushizima D M, Medeiros F N S, et al. SAR imagery segmentation by statistical region growing and hierarchical merging[J]. Digital Signal Processing, 2010,20(5): 1365-1378.

[4] Liu Bo, Cheng H D, Huang Jianhua,et al. Probability density difference-based active contour for ultrasound image segmentation[J]. Pat? tern Recognition, 2010, 43(6): 2028-2042.

[5] Zheng Ying, Li Guangyao, Sun Xiehua,et al. Geometric active contours without re-initialization for image segmentation[J]. Pattern Recog? nition, 2009,42(9): 1970-1976.

[6] Zhang Kaihua, Zhang Lei,Song Huihui,et al. Active contours with selective local or global segmentation: A new formulation and level set method[J]. Image and Vision Computing, 2010,28(4): 668-676.

[7] Campbell N D F,Vogiatzis G, Hernández C,et al. Automatic 3D object segmentation in multiple views using volumetric graph-cuts[J]. Im? age and Vision Computing, 2010, 28(1): 14-25.

[8] Boykov Y, Funka-LEA G.Graph Cuts and Efficient N-D Image Segmentation[J]. International Journal of Computer Vision, 2006,70(2): 109–131.

[9] Pundlik S, Woodard D, Birchfield S. Iris segmentation in non-ideal images using graph cuts[J]. Image and Vision Computing, 2010,28(12): 1671-1681.

[10] Plataniontis K N, Venetsanopoulos A N. Color image processing and applications[M] .Heidelberg: Springer. 2000.

[11] Zhang Chi, Wang P. A New Method of Color Image Segmentation Based on Intensity and Hue Clustering Pattern Recognition[C]//2000. Proceedings. 15th International Conference 2000,3(3-7): 613-616.

[12] Ito S, Yoshioka M, Omatu S, et al. An image segmentation method using histograms and the human characteristics of HSI color space for a scene image[J]. Artif Life Robotics,2006(10):6–10.

[13] Tian Xianzhong, Hu Tongsen.A Method of Debris Image Segmentation Based on SVM[C]// IEEE Proceedings of the First International Multi-Symposiums on Computer and Computational Sciences,.2006: 307-311.

[14] Fan Xing, Zhang Guoping, Xia Xuezhi.Performance Evaluation of SVM in Image Segmentation[C]//IEEE International Workshop on Se? mantic Computing and Applications,2008:160.

[15] Mohamed R, El-Baz A, Farag A. Remote Sensing Image Segmentation Using SVM with Automatic Selection for the Kernel Parameters[C]//2005 8th International Conference on Information Fusion. 2005,2: 8.

[16] Geman S, Geman D. Stochastic relaxation, gibbs distributions, and the bayesian restoration of images [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1984, 6(6): 721-741.