Novel edge detector

Abstract

A major dilemma in edge detections is the choice of optimum threshold which lacks generality. The problem is further amplified by the presence of false edges in the image due to noise. Addressing this dilemma the paper presents a novel technique by exploiting scale correlation with in wavelet subband for two dimensional signals with a view to retain structural information. The image is decomposed by dyadic wavelet transform up to 4 th level through multilevel wavelet decomposition. The detail coefficients in concordant bands are multiplied after interpolation and then synthesized. Quartic root of resultant product yields edge map of the image coupled with noise suppression. Experimental results reveal that the proposed algorithm outperforms the classical edge detectors for real, synthetic and noisy images while it is simpler to implement. © 2008 Springer-Verlag Berlin Heidelberg.