Research On Multi-exposure Fusion Based On Multi-scale Detail Enhancement

The dynamic range of images captured by ordinary digital cameras cannot meet the urgent needs of people for high-definition images.The emergence of multiple exposure image fusion technology can effectively solve this problem.The multi-exposure image fusion algorithm achieves the purpose of expanding the dynamic range of the fused image by fusing detailed information of different exposure intensity images.Multi-exposure image fusion is divided into two categories according to the shooting scene.One is the problem of loss of detail and halo in the fusion image in a static scene,and the other is the problem of ghosting in the fusion image in a dynamic scene.Aiming at these two kinds of problems,a multi-scale detail enhancement and ghost removal algorithm by studying the theory of multiple exposure image fusion is proposed in this thesis.First,the multi-scale fusion framework is used to remove the halo phenomenon in the fused image.Then,the edge-preserving filtering smoothing weight is used to retain the image texture information and an adaptive detail enhancement algorithm is proposed to enhance the image detail information.Finally,a ghost detection and elimination algorithm based on two-dimensional information entropy is proposed in this thesis,which can effectively detect and eliminate ghosts in the fused image.The main work of this thesis includes:1.Aiming at the problem that the existing multi-scale fusion framework easily leads to the loss of image edge texture information,the Gaussian weight pyramid is smoothed by weighted least squares filtering in this thesis.Weighted least squares filtering is an edge-preserving filter that can prevent relative brightness changes in the fused image,and effectively retain the image edge texture information,resulting in a fused image with a clear texture and rich details.2.Aiming at the problem that the detail information of the fusion result is not clear in the traditional multi-exposure image fusion algorithm.an adaptive image detail enhancement method is proposed in this thesis.First,obtain high-frequency detail information which reflects changes in the details of the fused pyramid image.Then,the image gain factor is designed,and the high-frequency component of the image is adaptively enhanced by the gain factor.Finally,laplacian reconstruction is used to obtain the fused image with enhanced detail.3.To solve the problem of ghosting caused by the direct fusion of images in dynamic scenes.a ghost removal algorithm based on two-dimensional information entropy is proposed in this thesis.First,the histogram matching algorithm is used to adjust the exposure intensity of the non-reference image to make it consistent with the exposure of the reference image,and the difference image is calculated.Then,the two-dimensional information entropy of the difference image is used to divide the dynamic and static pixel regions to obtain ghost-free weights.Finally,the ghost-free weight is used as the weight measurement factor,and the initial weight map is constructed by combining the remaining three weight items,and the weighted fusion is performed under the improved multi-scale fusion framework to obtain a ghost-free fusion image.