Low Snr Infrared Image Preprocessing And Segmentation Techniques

Object segmentation is a key technology in the infrared image processing system. It has important practical significance to segment the object accurately and effectively, especially for the target detection, identification, and tracking, etc. Usually, the infrared images have large noise, complex background, and ambiguous target. So in this dissertation the source and characteristics of the noise are studied, and the pre-processing and segmentation approaches are combined together in order to achieve a good segmentation result.There is a brief description of three parts about infrared image, which mainly are objectives, noises, and the background. At the same time, its corresponding mathematical model is established. On the basis of mastering the theoretical knowledge of the main source and characteristics about noises, the pre-processing algorithms are studied. The infrared image preprocessing technology needs to analysis features about a variety of filters (E.g.: smoothing filtering, frequency domain filtering, and morphological filtering). Then the images are filtered by the way of simulation on Matlab, and then we choose the appropriate pre-processing algorithms in the simulation results. At last, Simulation results show that the Median filtering and Morphological filtering can achieve better results. The background noises are suppressed in some extent, while the edge details of the target are maintained more completely.As the images preprocessing are done, the research about segmentation algorithms are began to study. Threshold segmentation methods are summed up and some algorithms such as Otsu, Bimodal method and Iterative method are simulated on Matlab platform. Then edge detection techniques are discussed in detail, which the serial and parallel boundary detection technologies are included. Among these algorithms, differential operators for edge detection (E.g.: Roberts, Canny, Sobel, Prewitt) and non-linear Laplace are particularly simulated on Matlab. Above all, the test results of two types of algorithms gotten by simulation are analyzed, and some good suggestions are adopted in the next improved segmentation methods. Three improved segmentation algorithms are proposed, such as: the method of combining the improved Canny multi-edge detection and simplified Otsu, the improved segmentation of edge detection by Interpolation method, the combination of morphological segmentation with the improved largest fuzzy divergence between two classes. Finally, the experiments results prove quite good in simulation.