Research On Image Fusion And Depth Reconstruction Based On Multi-focus Images

With the continuous development of digital image processing technology,the image contains more and more diverse information about the shooting scene.However,due to the operating principle of the image acquisition device and the application characteristics of different scenes,the information about the scene contained in the pixel values in the single two-dimensional image is always limited.The multi-focus image fusion is used to process a plurality of source images focused on different positions in the same scene,and extracts clear pixel information in each image to obtain a clear image.The application of multi-focus image fusion can greatly improve the utilization of image information,and provide complete scene data for subsequent image processing applications.In addition,multi focus images contain depth information of scene in the process of focusing plane changes.The depth reconstruction of the scene based on the depth information and the illumination color information can obtain a three-dimensional structural model containing more details of the scene.In this thesis,the problem of image fusion and depth reconstruction based on multi-focus images is studied,and new improved methods are proposed.Firstly,the application background of this thesis is microscopic scene.The image fusion and depth reconstruction algorithms proposed in this thesis are based on a zoom three-dimensional microscopic imaging system.This thesis analyses that the image acquisition environment of the system has some interference factors such as the change of transmittance and jitter.Aiming at these characteristics,the necessary pretreatment methods such as brightness balance adjustment,image scaling and registration are given.The contrast sensitivity model of human visual system is applied to image filtering.After these preprocessed methods,the original images have high availability.Secondly,combining the advantages of two methods to measure image clarity from different perspectives but with sufficient data features,a hybrid contrast factor is proposed to generate the decision results for each frame of multi-focus image in each local position.The decision result is weighted by Laplace energy to obtain the fusion image.The experiment compares the fusion results of various image fusion methods on multiple sets of data,and then quantitatively analyses the quality of fusion results of each method by various objective quality evaluation methods of fusion images.Experiments show that the quality of image fusion method proposed in this thesis is high and the computation time is moderate.Then,a depth reconstruction method based on structural similarity is proposed.The focus degree is measured by the similarity between each frame of multi-focus images and the fused image,and then the frame sequence number corresponding to the best depth value is calculated by the distribution weight of focus degree.Aiming at the situation that the result of depth reconstruction is distorted in high noise,a depth value restoration method based on anisotropic diffusion is presented.In this method,the area to be repaired is determined innovatively by peak signal-to-noise ratio of focusing degree,and then the depth value of the repaired area is calculated by anisotropic diffusion theory,so that the result of three-dimensional reconstruction have even higher accuracy.Experiment compared multiple depth reconstruction methods based on multi-focus images,and verified that the depth reconstruction method based on structural similarity was more accurate.Finally,the content of this thesis is summarized,and the direction of further research and improvement are pointed out about the structural design of the zoomable 3D microscopic imaging system,image fusion and depth reconstruction algorithms.