Fusion Of Infrared And Visible Images

The fusion of infrared and visible images is a very important part of the field of image fusion. Its applications spread in the areas include the military, security monitoring and so on. The imaging principle of the infrared sensor and visible sensors is different. Infrared imaging is a passive imaging, determined by the distribution and emission rate of the temperature of the object, the infrared image intensity distribution is mainly determined by the radiation changes of the background objects, and the contrast and resolution is generally low. However, the advantages of infrared image is that under certain environment conditions, such as smog, dark or object obscured, infrared image can effectively generate the images of the objects. Therefore, the information of the infrared and visible images can be treated complementary. In order to obtain more accurate and detailed information of the scene, fusion of infrared and visible images is used to make full use of the two kinds of image information to make better decisions.This thesis studies some key techniques of infrared and visible images fusion, including the following three aspects:First of all, the feature point extraction. We focus on corner detection. Compared to visible image, infrared image has low contrast, poor the target surface texture detail reflection, the blurring effect of edge and other features. The effect of using the Harris corner extraction operator on the IR image is not satisfactory. If the threshold is too large, some corners will not be extracted while reducing the threshold will cause many false alarms. In such cases, sub-block method is using to extract corner points. Entropy and correlation of images is used to limit the threshold of the Harris method ensuring that the the corners detected are true positives which will benefit the following processes.Secondly, on the basis of membrane optimization algorithms with hierarchical structure and the feature of the point set matching problem, a novel point set matching algorithm is proposed in this thesis. Matching relationship between the points are seeing as the material within the membrane region. To reduce the complexity of the algorithm, a kind of simulated annealing algorithm is used as a sub-algorithm for material evolution. In the proposed algorithm, three kinds of new heuristic search rules are introduced, by which matching rate is increased to some extent.Finally, an improved Laplacian pyramid algorithm is proposed. Different fusion rules are used for image sequence. A weighted fusion method based on genetic algorithm optimization is used on the top-level and the rule based on regional energy is used on remaining layers. The experimental results show that the method can be effective to retain the image information.