Radar Images And Optical Image Fusion Algorithms

In this highly developed information age, researchers hope to gain more useful information from the limited resources, to meet the needs of a higher standard. Since image plays a vital role in carrying information, cumulatively image-processing methods are efficiently carried out in the various domains of communication. To convey more information, image fusion is a very basic method.In this paper, a new approach based on the nonsubsampled contourlet transform (NSCT) is presented. The proposed method employs an adaptive fusion algorithm using the high frequency informations of radar image and optical image. The paper is structured as follows. In section1, image pre-processing is described. To ensure the features of the radar and optical images, Median filtering and Gaussian smoothing are adopted repectively. After denoising, the two types of images have to be registrated for further processing which is discussed in section2. In section3, the criterion of image fusion algorithm is introduced. The resulting image should be more informative and it is evaluated in section4. There are two elements of originality in this paper. One is about the capture of the implicit linear information in the image using the Hough transform, which is for further registration of the previously segmented images by longitude and latitude. The other one is based on NSCT fusion method. In different decomposition layers or in the same layer with different directions, fusion rules are adaptive for the high-frequency information of images. The ratio values of high frequency information entropy, variance, gradient and edge strength are calculated after NSCT decomposition. Experimental results showed that our approach performs better than those methods with single fusion rule.The experimental result in this paper demonstrates that new image fusion algorithm based on NSCT performs better.while the computing for the fusion is rather slow. In addition.for images without obvious features, the fusion produces undesirable result. Therefore, it worth researching to how to improve the registration accuracy and raise the algorithm efficiency in the future.