Abstract: Traditional image edge detection algorithms do not perform well in noise suppression. For
solving the problem, this paper presents a new method based on wavelet and local gradient field
singular value decomposition to detect image edge. First, the wavelet transform (WT) is applied to the
image in order to utilize the time-frequency localization characteristic of WT to extract the accurate
edge character. Then the local gradient field singular value decomposition (SVD) is used to the
gradient field calculated by the wavelet. The feature and good stability of the singular values (SVs)
enhance edge character and suppress noise. The experimental results of edge detection illustrate that
the proposed method can not only extract clear and complete single-pixel edge in the images without
noise, but also find ideal edge from noised images.

Key words: edge detection, wavelet transform, singular value decomposition, noise suppression