Edge Detection by Wavelet Scale Correlation

The spatial and scale space domain techniques are used independently to detect edges of the noisy images. When noise density surpasses a limit, classical operators are unable to detect the edges. The frequency domain filtering for edge detection in a noisy scenario is inadequate due to Fourier’s global behavior. Wavelet analysis for noisy images also reveals dominance of noisy pixels over the edges. Even multiresolution analysis falls short to distinguish noise and edge points in the synthesized image for depleted signal to noise ratio. In this paper noisy images have been decomposed up to fourth level through multilevel wavelet decomposition. The wavelet details coefficients are thresholded by four times the mean value of the image matrix. The lower dimensional wavelet detail’s coefficient matrices are interpolated up to the original size of the image. The noisy pixels are partially eliminated at each scale. However in the process, few edge points are also deteriorated. Independently multiplying each detail matrix by its three higher scale image matrices respectively significantly reduces the noise and enhances the directional edges. The reconstruction results in enhanced horizontal, vertical and diagonal details. The three images are synthesized to obtain the augmented edge map of the image. Key-Words: Edge detection, Multiscale, Multiresolution, Wavelet, Scale correlation, image denoising.