Research On Refocusing Algorithm Of Light Field Camera

In recent years,the rapid development of optical field camera technology has attracted wide attention.Different from traditional cameras,light field cameras can capture light field information and obtain multiple images of a scene in different angles at the same time,that is,light field data from multi-dimensional perspectives can be obtained.However,due to the limitation of optical system,light field camera can only select a specific focus when shooting.The refocusing algorithm can process the light field data,so that the focus plane can be moved and adjusted in the later stage,so as to obtain the image information of multiple focal points on the same image.Therefore,the refocus algorithm plays an important role in the optical field camera.The resolution of the image directly obtained by the optical field camera is relatively low,so it is necessary to improve the refocus algorithm to improve the clarity and resolution of the image.In this study,a new refocusing algorithm is proposed for the image shot by the lightfield camera.The algorithm designs DBI Downsampling Bicubic Interpolation algorithm and Transformer feature extraction module.First of all,after the data preprocessing of the image data taken by Lytro camera,the accuracy and computational efficiency of the refocusing image are improved by using the down-sampling double-cubic interpolation algorithm.The main idea of this algorithm is to downsample the original image and simplify the 16 pixel points required by the original double-cubic interpolation algorithm into 9 pixel points by designing the sampling matrix.Thus reducing the amount of computation.In order to sample more accurate image features,the 9 pixels in regular distribution are changed into irregular points more consistent with features.Secondly,in order to overcome the problems of refocusing algorithm sharpening and artifact,Transformer feature extraction module is designed to solve the problem.Its principle is to obtain a set of feature graphs containing various spatial and frequency information of optical field images through a series of operations such as convolution,normalization and nonlinear activation function.Then,through the self-attention mechanism,these feature graphs are interacted and fused.Experimental results show that the image accuracy is improved and the computational efficiency is faster with the downsampling bi-cubic interpolation refocusing algorithm.Meanwhile,the downsampling refocusing algorithm integrated with Transformer can better retain the image details and texture information,and avoid sharpening and artifact problems while improving the image resolution,which has better effect and stability.The main contribution of this paper is to combine optical field camera technology with image processing technology,and propose a method combining sampling interpolation refocusing algorithm and Swin Transformer feature extraction module to solve the problem of image refocusing.This method not only has good effect,but also has high practicability and popularization value.In future studies,we can further explore the application and optimization of refocusing of light field cameras,as well as explore more efficient and accurate refocusing algorithms.