Image segmentation based on competitive learning

Abstract

Image segment is a primary step in image analysis of unexploded ordnance (UXO) detection by ground penetrating radar (GPR) sensor which is accompanied with a lot of noises and other elements that affect the recognition of real target size. In this paper we bring forward a new theory, that is, we look the weight sets as target vector sets which is the new cues in semi-automatic segmentation to form the final image segmentation. The experiment results show that the measure size of target with our method is much smaller than the size with other methods and close to the real size of target.

Keywords

image segment / competitive learning / GPR / UXO

References

[1] Lee Paik. ABIDI. Image processing-based mine detection techniques: A Review[J]. Subsurface Sensing technology and Applications, 2002, 3(3): 153-202

[2] HIGGINS W E, WANG A J, REINHARDT J M. Semi-automatic 4D analysis of cardiac image sequences. Http;//gehrig.ee.psu.edu/publications/paper/paper.html

[3] Rodriguez R. Blood vessel segmentation via neural network in histological image[J]. Journal of Intelligent and Robotic System, 2003, 36: 451-465

[4] Kobashi S, Kmiura N. Volume-quantization-based neural network approach to 3D MR angiography image segmentation [J]. Image and Vision Computing, 2001, 19: 183-193

[5] Kurgollus F, Sankur B. Image segmentation based on multi-scan constraint satisfaction neural network[J]. Pattern Recognition Letters, 1999, 20: 1553-1563

[6] Shen Ding-gang, Horace H. S, et al. A Hopfield neural network for adaptive image segmentation: An active surface paradigm[J]. Pattern Recognition Letter, 1997, 18: 37-48

[7] Lin Jzau-sheng, Cheng Kuo-sheng, Mao Vhi-wu. Segmentation of multispectral magnetic resonance image using penalized fuzzy competitive learning network[J]. Computers and Biomedical research, 1996, 29: 314-326