The Application Research Of Wavelet Transform In Image Edge Detection

Edge is the most basic feature of images, so edge detection is an important content ofimage processing. In the past decades, the rapid development of the theory of wavelet hasbrought new theory and method for image processing. As wavelet transform has goodlocal quality and multi-scale identity, it can satisfy the need of edge detection inmulti-scales. Detecting edge using wavelet transform is recognized an efficient way.This thesis first introduces several current widely used edge detection algorithm suchas Sobel, Roberts, Laplacian. The core idea of these algorithms is that the edge pointscorrespond to the local maximal points of original image's gray-level gradient. However,when there are noises in images, these algorithms are very sensitive to noises, and maydetect noise points as marginal points, and the real edge may not be detected because ofthe noises' interference. The general idea of edge detection using wavelet transform is:choose a kind of suitable wavelet function, use the function to transform images inmulti-scale, detect the wavelet transform module local maximum, use given threshold toavoid noises and gain the image edge. In this article, we bring forward an edge detect waybased on two –dimensional image wavelet decompose details, use three high frequencyponderances of wavelet decomposition to detect image edge, and we use self-adaptivethreshold methods to obtain threshold. These methods include: the K-means clusteringmethod, the mean absolute deviation (MAD), the median method and the mean method.We perform all experiments based on these methods under the MATLAB environment,the results indicate that these methods are effective. Moreover, we analysis the advantagesand shortcomings of these methods, and point out the suitable images for them anddirection of further work.