Research On Image Fusion Algorithms At Pixel-Level Based On Wavelet Analysis

Image fusion is combining images or image sequence signals that get from two or more sensors at the same time (or different time) into a new single image. This new image can't get from any single sensor. The purpose of image fusion is to reduce uncertainty. Because the wavelet transform has good local property in time and frequency fields, it is widely used in image fusion. The thesis focuses on image fusion algorithm at Pixel-Level based on wavelet analysis. The image fusion algorithms based on multi-resolution analysis such as Mallat wavelet transform, lifting wavelet transform and nonsubsampled Contourlet transform (NSCT) proposed here to the fusion image are researched. By the theoretical analysis and many fusion experiments, a series of valuable conclusions is presented. The main work can be summarized as follows:(1) The thesis depicts the background, meaning, present status and problems of image fusion. Also, some typical algorithms of image fusion at pixel-level and some existing objective fusion performance measures are summarized.(2) The thesis is concerned with image fusion based on Mallat wavelet transform. Image fusion rules in wavelet transform region are introduced and a new image fusion algorithm based on wavelet transform is presented. The algorithm can better obtain contour information from source images by adaptive regional energy at the low frequency coefficients; and can better obtain details of information from source images by weighted regional strength ratio at the high frequency coefficients. By employing different masks, the algorithm integrates the approximation coefficients and the detail coefficients differently according to the weights, which are calculated with the regional energy of the approximation coefficients and the regional strength of the detail coefficients. Fusion image greatly improved Edge-dependent Fusion Quality Index (EFQI) and Weighted Fusion Quality Index (WFQI).(3) Considering the limitations of the fusion methods using the traditional wavelet transform to process a large quantity of image data, such as high computational complexity, need more memory and difficult to realize on-line fast image fusion processing, a new image fusion approach based on lifting wavelet transform is presented to combine multi-source images at pixel-level. Aiming at the coefficients of low frequency and high frequency, this algorithm chooses a different rule to fuse the image. To the low frequency, the spatial frequencies based on the neighborhood add consistency check is elected as the fusion guide. And the absolute maximum based on detail coefficients is selected as the guide to the high frequency. Taking the mean gradient,standard deviation and the ratio space frequency error as criterions, experimental results demonstrate that the algorithm is very effective and is able to fuse multi-source images.(4) The traditional wavelet transform has two main disadvantages: lack of shift invariance and poor direction sensitive. But the NSCT is a complete shift-invariant directional multi-resolution image representation, and it can retain the same dimension of the images of different levels which are obtained by the NSCT, so it can get more information in different levels by appending method. Therefore a new image fusion algorithm is proposed based on the NSCT. After the NSCT of images, the adaptive regional energy as the regular for the low frequency subband coefficients, the high-frequency detail images in different levels are processed by different fusion rules. The absolute maximum based on detail coefficients as the regular for the high level high-frequency and the weighted regional strength ratio as the regular for the others levels high-frequency. Although the algorithm increases the computation, the presented method can acquire much better visual quality and objective indexes as the wavelet transform and Contourlet transform.(5) In the end, the research work in this thesis is summarized, and the problems and disadvantage of current research work are also illustrated. Moreover, the paper discusses the recommendations for further work as well as the prospect of image fusion.