Research On Edge Detection Method Based On Wavelet Transform

Edge is one of the most basic feature of images, which reflects the most part information of images. The edge detecting is one of the most basic problems and hot research directions in the image processing, which has been widely applied in engineering, military, medicine and so on. Many scholars had done an in-depth study and obtained a number of results. So far, many algorithms have been presented in edge detection field. But natural images often contain different levels of noise. In special domain the noise and edges reflected the big changes of gray , and in frequency domain reflected the high frequency of the images. That bring many problems to edge detection. Edge detection is still one of classical, but not solved, technology. Wavelet analysis is a new tool of time-frequency analysis based on Fourier analysis. It can effectively analyze signal singularity point by its good time-frequency local property and multi-scale characteristics. Using the wavelet multi-scale transformation can detect the edge and detail much better, it also can remove most of the noise, and has strong noise suppression capability.First,edge detection methods, status and applications were briefly described in this paper. Second, the theory of wavelet analysis was introduced and lay a theoretical foundation for the wavelet edge detection algorithm. Several classical edge detection algorithms are theoretical analyzed in detail and experimented with MATLAB, and analyze their advantages and disadvantages after comparing the simulation results. Third, the theory of wavelet edge detection is elaborated in this paper, and gets the simulation results with MATLAB. The edge detection results in various scales are displayed using wavelet multi-scale features, and analyze the wavelet multi-scale edge detection algorithm characteristics. Finally, for the problems of wavelet edge detection algorithm, we improved the B-spline wavelet edge detection algorithm. We choose the third order B-spline wavelet as the wavelet bases. In the use of wavelet modulus local maxima, for the threshold selection, we use method of blocking adaptive threshold. In order to further remove the noise, we set up a threshold of the chain length to remove short chain.The images containing different noise or containing different details have been detected those edges by traditional Canny algorithm and the improved algorithm. The experimental results and data proved our method is superior to traditional canny algorithm. This improved algorithm can remove most of noise, and has the ability of the noise suppression. The simulation results showed the major edges of the images were detected effectively and reduced false edges of the images. This improved algorithm optimized the tradeoff between the accuracy of edge detection and the ability of the noise suppression. But the improved algorithm had several limitations in this paper: it detected edges with undesirable effects when the images had a lot of noise or many details and it is not applicable to these images. Generally speaking, our method is an effective edge detection algorithm and is superior to traditional canny algorithm.