Multi-Sensor Image Fusion Based On Clustering With Expectation-Maximization

Input images either come from only one sensor or multiple sensors with respect to output images,image fusion technique could combine a group of input images into a pair of output images which contains information of scene present better performance than any a pair of input image.As far as three decades,image fusion has been an important technique which improves the quality of information showed in output images,also which has been the principal motivation for developing image fusion technique.In the paper,the algorithm and application of fusing the multiple-sensor images are discussed,meanwhile,the infrared and visible images which are members of the multiple-sensor images are to be analysis object.In the paper,firstly,background and meaning of research related to multiple-sensor images fusion are going to be discussed,and the current evolution of fusing the multiple-sensor images is introduced.Secondly,the paper shows significant algorithms of fusing the multiple-sensor images in the aspect of theory and instances,including algorithms of multiple-scale transform,algorithms of region feature and algorithms of combining multiple-scale transform with region feature.Due to above those essential algorithms with flaws,according to the object state with explicit and hidden within infrared and visible images,the paper develops a segmentation algorithm for capturing the objective region based on Expectation-maximation(EM)clustering with mathmatical model,and segmented image is going to be a decision map.Besides,paper develops a method that initialize the parameters for EM.For applying the fusion image into fitting the human visual perception,in the paper,decomposing the infrared and visible image with Non-Subsampled Contourlet Transform(NSCT)and Stationary wavelet transform(SWT)respectively acquires the low-frequency and high-frequency coefficients.Finally,according to guiding by decision map,combining the low-frequency and high-frequency coefficients of the infrared image with low-frequency and high-frequency coefficients of visible image obtains the fusion image.In the paper,there are two algorithms of fusing videos proposed based on image fusion,and paper applies the video fusion in object detection.Additionally,there are some different groups of infrared and visible images which come from some different kinds of infrared and visible video are experimental instances,and the results of fusion are analyzed with subjective and objective form respectively.The results of segment by clustering show that algorithms of the paper not deal with a sheet of image but also process a video.The consequences of fusion experiment illustrate that algorithms of the paper could adapt to fusing different groups of infrared and visible images.Additionally,algorithm of the paper presents a better performance than traditional algorithms in extracting the object of infrared images and fusing the background information of visible images.Finally,the paper develops a graphical user interface based on Matlab to show the experiments and emphasize the applicability of the algorithm.