Research On Image Fusion Algorithm Based On Multi-Scale Analysis

Pixel-level image fusion is an important branch of the field of information fusion.It extracts important information from multiple images on the same scene obtained from multiple image sensors or the same image sensor in different working states,and integrates them into one image,to achieve the purpose of comprehensive understanding the scene from a single image.Pixel-level image fusion has been widely used in the military and civilian fields.The image multi-scale analysis method can describe an image from multiple scales,and separately process these different scales features,which is consistent with the human visual characteristics.Therefore,image fusion algorithms based on multi-scale analysis have very important research significance and application value.This paper mainly studies the key technologies and problems in image fusion algorithms based on multi-scale analysis.First,we investigated the development history of image fusion and the current research status at home and abroad,expounded the imaging characteristics of commonly used image sensors,combed the basic theoretical framework of image fusion,and briefly analyzed the hotspots and difficulties.The image fusion quality evaluation system is summarized and its existing problems are analyzed.Secondly,the multi-scale analysis tools commonly used in image fusion are introduced,and their advantages and disadvantages are summarized.Aiming at the problem that the RGF's decomposition results lack the direction information of the image and NSST cannot extract the subtle features in the image,this paper proposes a new image multi-scale decomposition method combining the respective advantages of RGF and NSST,which can both extract the subtleties into the image Features,but also can extract the orientation information of the image.Then,aiming at the problems that the fusion results obtained by many current infrared and visible image fusion algorithms are not natural and the infrared targets in the fusion results are not prominent enough,we designs low-frequency fusion rules and highfrequency fusion rules in the previous multi-scale decomposition framework in this paper: Combining the RPCA decomposition model and fuzzy logic to obtain the fusion weights of the low-frequency components,making the target information in the infrared scene more prominent in the fusion result;combining the traditional " max-absolute" rule and the local saliency to fuse the high-frequency components,making the fusion result more natural and consistent with human visual perception.Finally,we proposes an image fusion algorithm based on feedback.It can determine which source image has more prominent features at each local position,and design fusion weights based on this.It can retain the significant information of each source image in the fusion result.Compared with many traditional algorithms,it can also adaptively adjust parameters according to the characteristics of each image under the guidance of objective indicators,which is a quality-driven method.This article designed a detailed experiment to compare the above two algorithms with several classic algorithms and latest algorithms in terms of subjective visual effects and objective indicators.At the same time,the two algorithms are also compared in the same way.Experimental results demonstrate that the two algorithms in this paper can effectively improve the quality of the fused image and retain a lot of saliency information in the source image.