Abstract
Shape matching has been extensively used in various fields. The local feature-based or global feature-based algorithms can hardly describe the shape comprehensively due to their inherent defects. Combining the local and global feature to describe the shape has become a trend. In this paper, an improved discrete curve evolution algorithm is proposed which combines the discrete curve evolution with the uniform sampling and achieves a better description of the shape contour. Three simple and intuitive multi-scale features which represent both the global and local features of shapes are designed from aspects of the spatial relationship of contour points, structural information of contour sequence, and shape geometry feature. A cyclic Smith-Waterman algorithm is introduced to solve local contour matching and starting point selection. Experimental results demonstrate that our proposed features are translation, rotation, and scaling invariant, and have good robustness to deformation. Retrieval accuracies of Kimia99, Kimai216, and MPEG-7 indicate that our method can bring out a better performance.
Highlights
As an important issue in image processing, shape matching has achieved fruitful achievements in various fields, such as target recognition [1], image retrieval [2], [3], and biometric recognition [4], [5]
PARAMETER SETTING The ratio of the area of simplified contour to the original contour is used as the stopping criterion of improved DCE (IDCE)
The number of contour sampling points of different shapes can hardly be guaranteed consistent, so the initial sequence length N is set to Linit = max(0.08L, 3) in the evolutionary process experimentally, L is the length of the current contour, the length threshold M is set to Linit + 3
Summary
As an important issue in image processing, shape matching has achieved fruitful achievements in various fields, such as target recognition [1], image retrieval [2], [3], and biometric recognition [4], [5]. The global feature-based algorithms usually analyze the spatial distribution of image pixels or feature points from a global perspective to construct feature descriptors, and achieve shape matching through the similarity between features, such as shape context (SC) [6], [7], Fourier transform [3], curvature scale space (CSS) [8], feature invariants [9], [10], geometric hash value [11], mapping by. The global features have the property of being insensitive to noise, but the matching performance on deformation and occlusion is not good due to the missing of local details. The local feature-based algorithms can deal with occlusion and deformation, but they are susceptible to noise. We proposed a shape matching algorithm based on multi-scale invariant features of the contour sequence.
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
