Research And Application Of Wavelet Transform In Two-source Remote Sensing Image Registration

With the increasing development of techniques of acquiring remote sensing images, remote sensing image registration has became a crucial step for image mosaic, change detection and target recognition in the area of remote sensing application. Due to the high spatial resolution of remote sensing images, if the whole image is applied in the algorithm, it is computationally expensive, resulting in spending too much time, and cannot meet the requirements of practical application. Therefore, research on developing fast, high-precision algorithms has became the hard topic in the filed of remote sensing image registration. This paper studies some algorithms of wavelet transform in two-source remote sensing image registration, by theoretical analysis and experimental results, the completed works are as follows:1. Introduce the basic theory of wavelet transform and the Mallat algorithm of two dimensional signal(that is image) in detail. Because of the defects of the discrete wavelet transform, we further introduce two kinds of improved wavelet transform: undecimated discrete wavelet transform(UDWT)and dual tree complex wavelet transform(DT-CWT), and finally analyze the principle of the application of wavelet transform in remote sensing image registration.2. An improved method of remote sensing image registration based on wavelet transform and Fourier-Mellin transform is proposed. Considering the advantages and disadvantages of Fourier-Mellin transform and maximum mutual information when they calculate the image transform parameters, the method introduces discrete wavelet transform, for the low-frequency sub image after decomposition, First Fourier-Mellin transform is used to calculate the transform parameters, then from the lowest to the highest resolution image, maximum mutual information is used to optimize the parameters layer by layer, util the completion of the registration. Experiments show that, Comparing with methods based on Fourier-Mellin transform or maximum mutual information, the algorithm can greatly reduce the calculation time and achieve more accurate results of two-source remote sensing image registration.3. Owing to SIFT can extract image feature points with strong stability and high distinction, A method of remote sensing image registration based on wavelet transform and SIFT is analysised. In order to overcome the defects of the discrete wavelet transform and get richer features of the original image, We propose two methods of remote sensing image registration based on the improved wavelet transform and SIFT. Experiments show that, Comparing with method based on wavelet transform and SIFT, method based on the UDWT and SIFT can obtain more detailed partial information of frequency domain of the images, which contributes to extract more accurate image features. As for two-source remote sensing images with rich edge features, method based on the DT-CWT and SIFT can extract more advantageous edge feature points, which makes more precision result in image registration.