Infrared Image Edge Detection Algorithm Based On Level Set

As a low-grade vision dispose process, edge detection is always one of the hot research problems in image processing and analysis fields. Research on edge detection has last for the past decades. During this period, there have appeared several classical edge operators, such as Roberts, Sobel, Kirsch, Log, and Canny. But these methods are ineffective for infrared images, since the characteristics of infrared images include high noise, small distinction between target and background, and blur edge.Level set method introduced by Osher and Sethian is a numerical method, which translates the problem of evolution of 2-D (3-D) close curve (surface) into the evolution of level set function in the space with higher dimension. Level set is an accurate and steady algorithm; it can treat with the change of topology naturally. For the unique computing mode, level set obtains many good features, and attracts more and more attention in the field of edge detection.This dissertation focuses on the theory of edge detection based on level set and its application in infrared images. The primary research content and innovation are as follows:1. The theory of level set and its relevant numerical calculation methods are introduced. Through summarizing the features of the classical edge detection methods and level set, the superiority of detecting blur image based on level set is explained.2. According to the characteristics of infrared images and the configuration characteristics of level set control equations for two types of models, the necessity of detection infrared images based on Mumford-Shah model is elucidated.3. The characteristic of intensity distribution in two classes of infrared images(including infrared images of PCB and porcelain cup) are analyzed in detail. The problems of detecting the two classes of infrared images based on C-V model are discussed.4. Contrapose to infrared images of PCB and porcelain cup, background homogenization Mumford-Shah model and target homogenization Mumford-Shah model are proposed. The principles of homogenization background and target, the schemes of preservation object edges, the method selection of detected regions and undetected regions, the stopping criterions, the acceleration techniques of algorithms, and the effect of the parameters on algorithms are discussed in detail. Compared with C-V model and the classical edge detection methods, the efficiency and superiority of the two models have been proved by the results in the simulated experiment.