Research On The Multi View Multi Exposure Image Fusion And Quality Assessment

Image fusion is the process of extracting useful information from different images and combining them into a high quality image.Multiexposure image fusion is one of the most important part in the field of image fusion.Influenced by brightness and exposure time,images may present under-exposure or over-exposure,sometimes even lose some important information.The research about multi-exposure image fusion is to mix multi-exposure images captured from same view.Considering the complexity of scenes,it is hard to ensure capturing images from the same view.So corresponding pixel matching of pictures may present some mistakes.Based on such scenes,our article will discuss the multi-exposure image fusion from different views.First we try to get features from multi-view images under the similar scales and make hypergraph model.We regard the brightness enhancement of these points as labels,and introduce random walks model.Then we try to get the probability of one pixel walking randomly to the seed points.Finally,we ascribe the walking points labels,and get final fusion result.Meanwhile,this article proposes a quality evaluation method to multi-view fusion image.People always use statistical-based average square root error method or other methods,but considering the distortion caused by multi-view images,traditional method is not suitable.Towards this problem,this article proposes the quality evaluation method based on local edge.We redefine the concept of edge,and regard the reserved edge from the sample to the result as the evaluation index.Finally,this article uses samples from Daniel Scharstein Database and our own samples in SJTU,and make samples fused.We get perfect result and look forward to the future research.