Research On The Multi-exposure Image Fusion And Quality Assessment Of The Fused Image

At present,the brightness range that the imaging sensor of an ordinary digital camera obtained is much smaller than the real scene and the human eye can perceive.Therefore,the single image can only display the image information within a certain brightness range of the shooting scene,this leads to the loss of details in the overexposed and underexposed areas of the image.Multi-Exposure Image Fusion Technology and High Dynamic Range(HDR)Technology can solve the above problems through the different exposure image processing in the same scene.The HDR technology uses the camera response function to synthesize the HDR radiometric maps,and compresses the dynamic range by the method of tonal mapping,then get a new image with the clear details of each area that can be displayed on a common device.However,the HDR technology has the problem that the camera response function is inaccurate and time-consuming.Multi-Exposure Image Fusion Technology directly on the multi-exposure sequence image fusion without the need for the camera response function,which runs faster,and is more suitable for use on mobile devices.In this paper,firstly,a static scene multi-exposure fusion algorithm based on image block component extraction is proposed.The image blocks are extracted into three independent components-the structure contrast,the structure direction and the average exposure,then these three components are fused separately.After that,the new fused image is obtained by reconstructing the fused components.In this method,the RGB three-channel information is used to fuse three components at the same time,and the color of the fusion image is brighter.For various reasons,multi-exposure image capture scenes often have the existence of moving targets,this leads to the existence of ghosting after the fusion of the static scene algorithm.Therefore,it is need to detect moving targets before the fusion.In this paper,on the basis of the static scene fusion algorithm,firstly,using the exposure function to select the reference image,and using the structural directional component to calculate the structural consistency indicators of the other exposure images and reference images.Then the Dense Sift introduced to describe the consistency indicators of compute subspace.Using the structural and spatial consistency indicators to eliminate the moving targets,the fused image obtained without the existence of ghosting.In this paper,the experiments are carried out in 20 dynamic scenes.Compared with 12 kinds of classical ghost removal algorithms,the algorithm proposed in this paper has advantages in arithmetic speed and performance.In the multi-exposure fusion image quality evaluation,academia has not formed a public opinion.A multi-exposure fusion image quality evaluation algorithm based on visual saliency proposed in this paper.First,the fusion image to be assessed is divided into image blocks,and the quality assessment model of the image blocks is established through deep convolutional neural network.The importance weight of image blocks is obtained by using the significance extraction algorithm based on regional contrast and Prewitt edge detection to evaluate the importance of different image blocks.The fusion image quality score obtained by using image block score weighted.Finally,the multi-exposure fusion image was quantitatively analyzed by this evaluation method,and the proposed multi-exposure fusion algorithm was automatically tuned.