A robust contour detection algorithm is presented for noisy images characterised by close objects. The proposed approach uses an adaptive multi-scale edge tracking scheme based on active shape models and the wavelet transform. This adaptive method effectively adjusts the appropriate Gaussian function bandwidth according to the noise level so that close object edges can be detected before they are merged by excessive smoothing. This gives an improved performance over a single scale approach, where an incorrect Gaussian function bandwidth can lead to erroneous edge detection. The results obtained show an adaptive multi-scale scheme is robust regardless of the image signal to noise ratio.
[1]
Sasan Mahmoodi,et al.
Automated vision system for skeletal age assessment using knowledge based techniques
,
1997
.
[2]
Stéphane Mallat,et al.
Singularity detection and processing with wavelets
,
1992,
IEEE Trans. Inf. Theory.
[3]
John F. Canny,et al.
A Computational Approach to Edge Detection
,
1986,
IEEE Transactions on Pattern Analysis and Machine Intelligence.
[4]
Timothy F. Cootes,et al.
Training Models of Shape from Sets of Examples
,
1992,
BMVC.
[5]
C. Taylor,et al.
Active shape models - 'Smart Snakes'.
,
1992
.