Structural edge detection using wavelet domain statistical model
Shubin Zhao · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2009
Edges play an important role in most computer vision and image analysis systems, and extracting structural edges is the main goal of edge detection for many applications. But due to the presence of noise and texture, structural edge detection is not a trivial work. In this paper, an approach is presented for structural edge detection, which is formulated as a statistical pattern recognition problem in wavelet transform domain. In the approach, both inter-scale and intra-scale dependences among wavelet coefficients are utilized, where the former dependences are encoded by inter-scale coefficient ratios and the latter by anisotropic MRF model. To reduce the computational complexity, independent mixture of Gaussian is used to model wavelet coefficients, which corresponds to Rayleigh distribution for gradient magnitude, and posterior probabilities are computed to measure the edge strengths. Inter- and intra-scale dependences among coefficients are utilized to suppress noise and texture, and these measures can significantly improve edge continuity in scale space. To show the effectiveness of the presented algorithm, experiments are conducted on various kinds of real-world images, and several results are given for assessment.