Multiscale Edge Detection Of Image Based On Lifting Wavelet Transform

Edge is the import feature of unregular structure and non-stationary signal in image, which gives the profile and usually takes lots of information, the most concernful feature of image recognition. As a result the first and basic stage in image processing is the edge detection, the result which will directly influence the accuracy of recognition.Due to the diversity of image, noise level, and the difficulty of extracting some intriguing edge from procured images, there are not a math model and uniform method until now. In this thesis, some new algorithms to detect image edge are studied and the major ones are as follows:Firstly, a new arithmetic based on Walsh-Haar type transform is proposed for the major problem in edge detection of image signal: the contradictions of smoothing and approximating. Although the edge of image is not single pixel for adding image fusion, however compared with other methods, this algorithm with which the edge image includes more information on the brink is simple.To improve the disability of extracting single pixel in the algorithm mentioned above, a new method of edge detection is presented based on lifting wavelet in this thesis, the simulation experiment is finished. The results show us that the method can be used to many kinds of wavelets with the advantages of higher speed and precision.