A new unsupervised color segmentation method based on region-constrained EM (Expectation Maximization) and graph cuts is proposed, which can automatically determine the number of segments for a color image.The proposed method first obtains the superpixels of the image and extracts CIE Lab color feature and multi-scale quaternion Gabor filter feature.In order to automatically and efficiently determine the number of segments for the image and avoid the problem caused by using superpixels directly, a window is used to sample each superpixel to obtain a pixel subset which represents the superpixel.Then the feature space of the sampled pixel subsets is modeled with Gaussian mixture model, and the model parameters (including the number of components) are obtained by a region-constraint component-wise EM algorithm.Finally, the segmentation result can be obtained by α-expansion with the learned model parameters.Experimental results demonstrate the good performance of the proposed method.
[1] Deng Y,et al.Unsupervised segmentation of color-texture regions in images and video[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2001,23(8):800-810.
[2] Yang A Y,Wright J,Ma Y,et al.Unsupervised segmentation of natural images via lossy data compression[J].Computer Vision and Image Understanding,2008,110(2):212-225.
[3] Mobahi H,Rao S R,Yang A Y,et al.Segmentation of natural images by texture and boundary compression[J].International journal of computer vision,2011,95(1):86-98.
[4] Comaniciu D,Meer P.Mean shift:a robust approach toward feature space analysis[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2002,24(5):603-619.
[5] 王顺凤,冀晓娜,张建伟,等.局部熵驱动的GAC模型在生物医学图像分割中的应用[J].电子学报,2013,41(12):2487-2492. WANG SHUN-feng,JI Xiao-na,ZHANG Jian-wei,et al.Application of GAC model driven by the local entropy on medical image segmentation[J].Acta Electronica Sinica,2013,41(12):2487-2492.(in Chinese)
[6] 赵荣昌,马义德.一种用于图像编码的区域分割新方法[J].电子学报,2014,42(7):1277-1283. ZHAO Rong-chang,MA Yi-de.A novel region segmentation algorithm with neural network for segmented image coding[J].Acta Electronica Sinica,2014,42(7):1277-1283.(in Chinese)
[7] Boykov Y,Veksler O,Zabih R.Fast approximate energy minimization via graph cuts[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2001,23(11):1222-1239.
[8] Nguyen T M,Wu Q J.Gaussian mixture model-based spatial neighborhood relationships for pixel labeling problem[J].IEEE Transactions on Systems,Man,and Cybernetics,Part B:Cybernetics,2012,42(1):193-202.
[9] Ji Z,Liu J,Cao G,et al.Robust spatially constrained fuzzy c-means algorithm for brain MR image segmentation[J].Pattern Recognition,2014,47(7):2454-2466.
[10] Chen S,Cao L,Wang Y,et al.Image segmentation by MAP-ML estimations[J].IEEE Transactions on Image Processing,2010,19(9):2254-2264.
[11] Yang Y,Han S,Wang T,et al.Multilayer graph cuts based unsupervised color-texture image segmentation using multivariate mixed student's t-distribution and regional credibility merging[J].Pattern Recognition,2013,46(4):1101-1124.
[12] Figueiredo M A,Jain A K.Unsupervised learning of finite mixture models[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2002,24(3):381-396.
[13] Yang Y,Guo L,Wang T,et al.Unsupervised multiphase color-texture image segmentation based on variational formulation and multilayer graph[J].Image and Vision Computing,2014,32(2):87-106.
[14] Li L,Jin L,Xu X,et al.Unsupervised color-texture segmentation based on multiscale quaternion Gabor filters and splitting strategy[J].Signal Processing,2013,93(9):2559-2572.
[15] Martin D,Fowlkes C,Tal D,et al.A database of human segmented natural images and its application to evaluating segmentation algorithms and measuring ecological statistics[A].Proceedings of the Eighth IEEE International Conference on Computer Vision[C].Vancouver,British Columbia,Canada:IEEE,2001.416-423.
[16] Kim T H,Lee K M,Lee S U.Learning full pairwise affinities for spectral segmentation[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2013,35(7):1690-1703.
2016, Vol. 44 
