Research On An Edge Detection And Tracking Algorithm Based On Wavelet Transform

Edge is a composition of the high-frequency energy in the image,which is also the most important features of the image. In the visual interpretation of the image or the computer automatic identification, to analysis and perceive objects profile is an important means or process. Then, image edge extraction has become the hot issues in machine vision research.There are many kinds of the edge detection methods now, such as Sobel, Canny and other operators,but most of them are based on the gradient algorithm. These operators are very sensitive to noise due to the derivative operation and the convolution scale. The detected image edges are contaminated by noises to eliminate difficultly by low-pass filtering. Wavelet analysis is a novel mathematical means, which has been Widely used in Signal or image processing, due to its features of denoising and multi-resolution. This thesis is aimed at raising a more practical edge detection and tracking method using wavelet transformation theory. Around the topic, wavelet analysis in the application of edge analysis, the main content and conclusions are as follows:(1) Introduced several major edge detection operators, and advantages and disadvantages of each;(2) Described systematically the wavelet transform theory and its feature of multi-resolution in image processing;(3) Based on the wavelet multi-scale edge-detection algorithm and technical processes, a image edge extraction method were raised, including the image non-down sampling decomposition, two decision gradient operator and multi-scale edge tracking, with better image edges.(4) Evaluated our algorithm and others by signal to noise ratio evaluation model, with the help of the Symmetric images and remote sensing images from Vision Laboratory Purdue University, and the results show that this method is excellent without noises, but slightly worse than the Canny operator when adding 5% uniform noise.