|
Image registration refers to the combining of two partially-overlapped images, which may be subject to rotation, translation, scaling, intensity and color changes, into one large image. In registering two images, feature points of two images are found first. The matching pairs are then obtained by comparing the two sets of features. However, for saving the computation time of finding the matching pairs, we apply the wavelet transform to perform a multi-resolution registration. An initial estimation of the matching parameters (rotation, translation and scaling) is computed in the lowest resolution. Then, a more accurate estimation of the matching parameters is carried out in the high resolution. By using a coarse to fine matching strategy, we fast achieve the image registration.Since image registration is able to put the overlap of two images on the same plane coordinate, we can apply it in image fusion and image restoration. On image fusion, we can make use of image registration to correctly locate two differently-focused images of the same picture on the same plane coordinate. We can use algorithm to search for the apparent parts of the two images and fuse them into a totally-clear image. On image restoration, as for two images - a clear one and a blur one - with partial overlap, we can also make use of image registration to find out the overlap between them. By means of the overlapped parts, we are able to find out the blurring function of images; therefore, the whole image can be restored.The experiments justify that multi-resolution image registration can not only be applied in various types of images but finish image registration quickly as well. Although the overlapped parts of two registered has changed in image fusion and restoration, yet we can combine them successfully and find out the image effects we want by algorithm.
|