A multiple point boundary smoothing algorithm
Section snippets
1. Introduction
Boundary analysis plays an important role in pattern recognition applications (Pavlidis, 1980). Due to noise, the boundaries of a digitized object are often too complicated for direct analysis. To overcome this problem, boundary data are often smoothed before feature extraction.
There are many approaches for boundary smoothing. One may use some preprocessing algorithms to remove noisy spikes and fill small holes, where the deleting and filling patterns are applied to the whole image (Suen et
2. One point smoothing algorithm
The boundary of an image is traced and represented by its chain code. The outer boundary is traced in the counterclockwise direction while the inner boundary is traced in the clockwise direction.
Suppose a boundary P consists of a sequence of N points,where pi+1 is a neighbor of pi(i = 0,1,…,N–1) and the index i is calculated modulo N. The coordinates of the ith point are denoted as (xi, yi) and its chain code is given bywhere the Freeman chain code in 8
3.1. Boundary coding
We first define a boundary primitive as a set of consecutive points which lie in a line. In a primitive, all the points except the last one have the same chain code. This chain code is defined as the direction of the primitive. A boundary can be decomposed into a sequence of primitives and each primitive is then recorded by a boundary code. Suppose that there are M primitives in the boundary, then the boundary code bc[i] corresponding to primitive i is defined as follows:
4. Feature extraction
After the one point and multiple point smoothing algorithms, a noisy boundary becomes much smoother than before. This will reduce redundant features significantly. In addition, the possible relations between neighboring boundary primitives have been reduced since the angle between two primitives is 135° only. This will facilitate the feature extraction process.
Three kinds of features are extracted: the main feature vectors for the even chain code directions; the secondary feature vectors for
5. Experiments
We use handwritten numerals extracted from the NIST database SD3 (Garris and Wilkinson, 1992) to test the smoothing and feature extraction algorithms. The database consists of 2100 forms which contain handwritten samples of isolated digits, digit groups, English letters, and English prose. Images of isolated digits from the forms were segmented, manually classified. A total of 159 122 isolated digits was used in the experiment. The one point smoothing algorithm is applied to each boundary of
6. Conclusion
The one point smoothing algorithm by Yu and Yan (1997) provides an efficient way to smooth a noisy boundary. In this paper we have summarized this algorithm with four rules, and the effects of the four rules to the boundary are analyzed. Based on this algorithm, a multiple point smoothing algorithm has been proposed to smooth the boundary further. With this algorithm, the quality of a noisy boundary can be improved significantly, and thus redundant features can be reduced.
Three kinds of feature
Acknowledgements
The authors wish to thank the anonymous referees for their valuable comments. This work is supported by the Australian Research Council.
References (14)
- et al.
Corner detection for chain coded curves
Pattern Recognition
(1995) - et al.
Structural classification and relaxation matching of totally unconstrained handwritten ZIP-code numbers
Pattern Recognition
(1988) - et al.
An efficient algorithm for smoothing, linearization and detection of structural feature points of binary image contours
Pattern Recognition
(1997) - et al.
Syntactic recognition of handwritten numerals
IEEE Trans. Systems Man Cybernet.
(1977) - Garris, M.D., Wilkinson, R.A., 1992. Handwritten segmented characters database. Technical Report Special Database 3,...
- Hu, J., Yan, H., 1998. A model-based segmentation method for handwritten numeral strings. Comput. Vision Image...
- et al.
A stroke width compensation algorithm for character recognition
Electronics Lett.
(1996)
Cited by (10)
Thinning combined with iteration-by-iteration smoothing for 3D binary images
2011, Graphical ModelsCitation Excerpt :Since unwanted endpoints are salient object points, the proposed topology preserving smoothing algorithm is to remove additive contour noise elements. There exist numerous approaches for smoothing binary objects in 2D and 3D [15–18]. Yu and Yan developed a 2D sequential boundary smoothing algorithm that uses operations on chain codes [15].
Image analysis and computer vision: 1998
1999, Computer Vision and Image UnderstandingConstruction of partitioning paths for touching handwritten characters
1999, Pattern Recognition LettersAn intelligent video security system using object tracking and shape recognition
2011, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)From CT to NURBS: Contour fitting with B-spline curves
2011, Computer-Aided Design and ApplicationsTopology preserving parallel smoothing for 3D binary images
2010, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)