Image Fusion New Algorithm Based On The Combination Of NSCT And Region Segmentation

In recent years,image fusion has become an important and useful technique for image analysis and computer vision. Image fusion can produce a more precise, integrated and reliable estimation and description of them than a single image. For the characters of infrared and visible images, this thesis focuses on image fusion based on region segmentation and edge detection. The main contributions can be summarized as follows:Basic concepts of image fusion are introduced and many traditional image fusion algorithms are systemically analyzed. The theory of nonsubsampled contourlet transform(NSCT) are discussed and a new image fusion algorithm based on NSCT and edge detection are proposed. Firstly, multiscale decompositions of the source images are carried out with the NSCT. Then, the low-frequency coefficients are fused with the regional character, and the high-frequency coefficients are fused with the edge character and normalized correlation moment. Finally, the fused coefficients are reconstructed to obtain the fusion image.A new image fusion algorithm based on NSCT and region segmentation was proposed too. Firstly, graph-based and fuzzy c-means clustering(FCM) algorithm are applied for infrared and visible images segmentation respectively, and region relating segmentation images are received by region combination. Then, multiscale decompositions of the source images are carried out with the NSCT. The low-frequency coefficients are fused with the regional characteristics measurement operator of the ratio of region energy and the ratio of region sharpness, and the high-frequency coefficients are fused with the regional characteristics measurement operator of region edges clarity.The fusion experiments for several infrared images and visible images are taken, and comparisons with the local window energy algorithm based wavelet transform and nonsubsampled contourlet transform are given. The results show that the fused images obtained by the two proposed algorithms can not only hold plentiful background information of the visible image, but also inherit the object information of the infrared image.