Research On Key Techniques Of Multi-Source Image Fusion Based On Gradient Field

Image fusion is intersection between image processing and information fusion,while multi-modle image fusion is an important part of image fusion.Images captured by different sensors are usually greatly different from each other,which is mainly caused by different imaging principles of different imaging sensors.Multi-source image fusion can provide more information since differences between different image sources,but on the other hand,those differences lead to difficulty when applying single-source image fusion alorithems to multi-sources image applications.The core issue in multi-source image fusion field is to exhibit salient information that comes from different sources accurately at final fused image as much as possible in a way that is suitable to human visual system.Image gradient can effectively represent salient changing information such as edges or texture in any image source and this changing information is sensitive to human visual system.With help of image gradient,saliten information in different image sources can be effectively fused together.Therefore,there is much necessity and value to research gradient field based multi-source image fusion algorithems,and gradient-based algorithems are important parts of state of art multi-source fusion algorithems.This paper mainly focus on key technologies of gradient field based multi-source image fusion.Firstly,summaries of image fusion backgrounds,developments and state of art are made.Then imaging sensors that widely used in muti-source image fusion are analyzed about their imaging principles and fusion requirements.The application values and current existing problems are also mentioned in this paper.Then,gradient field building and reconstruction theories of digital image signals are introduced and derived in detail,which makes sure feasibility of gradient field based image fusion.Among many gradient based algorithems,structural tensor based methods are recently popular and effective.This parper introduces concepts about structural tensor and image fusion process based on that in detial.Disadvantages and advantages of structural based methods are also analyzed.On this basis,a local structural similarity metric is proposed firstly to measure local similarities betweent different image gradients.Then a gradient filter with adaptive scales is designed based on this similarity metric to adjust fused gradient field that obtained by structural tensor based method.This process will refined fused gradient to increase saliency,robustness and anti-noise ability of fusion process.Finally,abundant expriments are designed and excuted to test feasibilities of components in proposed algorithm one by one,more-over comparison between this paper and many classical and recent related fusion methods are carried out from many aspects such as: subjective,objective,detail,robustness,and anti-noise performance.Expriment results indicate that proposed local structural similarity based gradient filtering method can effectively improve quality of final fusion result while providing more information and having better anti-noise ability.