Research Of Image Registration And Fusion Technology Based On The Wavelet Transform

With the rapid development of science technology, more and more sensors are used in industry andeveryday life. But information in images generated from a single sensor is limited, thus it isoften insufficient for actual requirement. Image fusion is to utilize information acquiredthrough various sensors with different imaging methods. It could provide complementaryinformation so that more reliable and accurate images can be generated for observing andfurther processing. Image fusion is now widely used in military, medical science, remotesensing, computer vision and object detection etc. Image Fusion based on Multi-ResolutionAnalysis is a method according with visual effects, which can achieve good image quality butis often with redundancy. However the wavelet analysis is an implementation which has thedirectional and non-redundant characteristics, so it is superior. Wavelet Transformationbecomes a mainstream technology in image fusion at present time. However, existing imageregistration and image fusion methods based on wavelet transformation have severaldeficiencies. More accurate and efficient Methods are needed. Contributions of this thesis areas follows:(1)Image registration and fusion based on pixel level is introduced, including concepts,basic models and the classification of image registration, level division of image fusion andevaluation criteria of image fusion. Besides, advantages and flaws of common methods ofimage fusion are given. In addition, the related concept of wavelet transform is introduced.(2)For image registration, morphological open operation is used to eliminate noiseinterference first of all. Then, regional mutual information is used to take spatial informationbetween images into account. Its formula is improved, so that redundancy and computingcomplexity is reduced and that spatial information between images is taken into consideration.Moreover, phase consistency is used only when image resolution is high, so that matchingspeed is increased and a side effect caused by brightness variance between images iseliminated. Besides, a multi resolution hybrid optimization algorithm is utilized, which avoidslocal extremes and improves the image registration precision. The top floor which has lessdata adopts particle swarm optimization optimal registration method. Then the search result is used as the starting point of Powell optimization method in the next layer. In the end, thesimulation experiments show that the improved algorithm achieves higher precision, sub pixelprecision, and high robustness against noise of image.(3)For Image Fusion, original images are decomposed into sub images by using waveletdecomposition. A new method based on energy and related coefficient is proposed. The highfrequency sub images are fused according to energy contribution rate by pixels within thewindow. The low frequency sub images are fused by the related coefficient strategy. In theend, experiments carried out in MATLAB9.0show that the proposed method can get fasterspeed and higher image quality than previous methods.