Image segmentation based on complexity mining and mean-shift algorithm

Mean shift algorithm is a well established method for image segmentation. It is particularly popular technique due to non-parametric nature which enables efficient segmentation of complex arbitrary shapes. Despite such advantage, high computational complexity still makes it unsuitable for segmentation of high resolution images in time critical applications. This paper introduces a new approach which alleviates performance issues of mean shift using complexity reduction based on information theory. Proposed algorithm starts by calculating information potential field of the image in order to get insight into complexity of the regions. Afterwards, only complex regions are segmented by computationally expensive mean shift algorithm, while segmentation of simpler regions is performed by a cheaper method. Performance of our method is additionally improved with execution of the key code sections on the GPGPU platform. Experimental results have shown that our method produces comparable segmentation quality to regular parallel mean shift, but with significant reduction in overall execution time.

[1]  Balázs Varga,et al.  High-resolution image segmentation using fully parallel mean shift , 2011, EURASIP J. Adv. Signal Process..

[2]  Miguel Á. Carreira-Perpiñán Acceleration Strategies for Gaussian Mean-Shift Image Segmentation , 2006, 2006 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'06).

[3]  José Carlos Príncipe,et al.  Mean shift: An information theoretic perspective , 2009, Pattern Recognit. Lett..

[4]  Bedřich Beneš,et al.  Connected Component Labeling in CUDA , 2011 .

[5]  Dan A. Simovici,et al.  Entropy Quad-Trees for High Complexity Regions Detection , 2011, Int. J. Softw. Sci. Comput. Intell..

[6]  Yizong Cheng,et al.  Mean Shift, Mode Seeking, and Clustering , 1995, IEEE Trans. Pattern Anal. Mach. Intell..

[7]  Ming Tang,et al.  Accelerated Convergence Using Dynamic Mean Shift , 2006, ECCV.

[8]  Larry D. Hostetler,et al.  The estimation of the gradient of a density function, with applications in pattern recognition , 1975, IEEE Trans. Inf. Theory.

[9]  Miguel Á. Carreira-Perpiñán,et al.  Gaussian Mean-Shift Is an EM Algorithm , 2007, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[10]  N. Otsu A threshold selection method from gray level histograms , 1979 .

[11]  V. Papic,et al.  Accelerating mean shift image segmentation with IFGT on massively parallel GPU , 2013, 2013 36th International Convention on Information and Communication Technology, Electronics and Microelectronics (MIPRO).

[12]  Larry S. Davis,et al.  Mean-shift analysis using quasiNewton methods , 2003, Proceedings 2003 International Conference on Image Processing (Cat. No.03CH37429).

[13]  Ewa Skubalska-Rafajlowicz Local Correlation and Entropy Maps as Tools for Detecting Defects in Industrial Images , 2008, Int. J. Appl. Math. Comput. Sci..

[14]  Bin Fang,et al.  A Computational Model for Saliency Maps by Using Local Entropy , 2010, AAAI.

[15]  A. Rényi On Measures of Entropy and Information , 1961 .

[16]  Wu-chun Feng,et al.  To GPU synchronize or not GPU synchronize? , 2010, Proceedings of 2010 IEEE International Symposium on Circuits and Systems.

[17]  Dorin Comaniciu,et al.  Mean Shift: A Robust Approach Toward Feature Space Analysis , 2002, IEEE Trans. Pattern Anal. Mach. Intell..

[18]  Meng Liu,et al.  Efficient Mean‐shift Clustering Using Gaussian KD‐Tree , 2010, Comput. Graph. Forum.

[19]  Peter Meer,et al.  Synergism in low level vision , 2002, Object recognition supported by user interaction for service robots.

[20]  Deniz Erdogmus,et al.  Generalized information potential criterion for adaptive system training , 2002, IEEE Trans. Neural Networks.