Image Fusion Technology Based On Multi-scale Analysis

Image fusion is a process of combining different information from images of multiple sensors acquired a scene in same time, extracting the complementary information to construct a more abundant、more accurate and more reliable fused image. Image fusion is a very active research field, and it plays an important role in medical science, measurement,geographic information systems, industrial and intelligent robot and military areas. Recent years,researchers pay more and more attention to image fusion.This paper mainly studies the multi-sensor image fusion technology based on the multi-scale analysis. These studies focus on enhancing the quality and shorting the fusion time. According to difference research priorities, this paper studies systematically for two key points-extraction method and fusion rules. Some fusion methods are proposed in this paper,and large number of experiments proved the effectiveness of the methods. The concrete work is shown as follows:We gave the concept of multi-sensor image fusion and the development of multi-sensor image fusion technology, analysis characteristics of the existing methods and discussed the advantages as well as disadvantages, and for some drawbacks, proposed improvement. Finally,proposes a practical fusion method in real-time application.First, summarized the development situation of multi-scale image fusion, and give a brief introduction to the different sensors, such as Remote sensing image fusion, Medical image fusion, Multi-focus image fusion, and introduce the focus of every kinds of the image fusion.Presents a framework of visible and infrared image fusion, the image preprocessing image enhancement, and color space conversion are introduced in detail. Then, fusion evaluations is introduced, Q factor and structural similarity is introduced in detail.Secondly, this paper introduces the existing multi-scale image fusion methods, and the simulation test is given for all algorithms. On the comparison between continuous wavelet transform and discrete wavelet transform in fusion results and time consuming. According to the lifting framework of D9/7, the lifting frame of the D5/3 is put forward. Briefly introduces the CT transform.Then, detailed analysis and abundant experiments is done to search for methods how to improve the integration quality. Analysis the advantages and disadvantages of wavelet transform, lifting wavelet transform, contourlet transform and nonsubsampled contourlet transform. In nonsubsampled contourlet transform, the frequency band filter module would loss some detail information, for this problem, an improved method was put forward. The modified redundancy lifting transform was taken to replace the pyramid transform for image band decomposition. Experiential results shows that the proposed algorithm is better in maintain the visible details and get a clear hot object of the infrared target information.Finally, in the field of infrared and visible image fusion, secondary fusion method is proposed based on morphology, this method greatly shortens the fusion time and satisfy the requirement of real-time processing. The process of this method as follows: using top-hat transformation to extract the brightness information of the original images and obtained the first fusion image, as a result, the hot object information in infrared image would be extracted.Then, TCO is used to extract the texture features of the original images and the second fusion image would be obtained, as a result, the detail information in visible image would be extracted. And then, weighted average method is used for quadratic fusion. Experiments show that the quadratic fusion method with high-quality integration and short fusion time, it meets the requirement of engineering practice.