Estimation Of Light-field Depth Flow Based On Variational Method

Different from the traditional images which can only record the light intensity information and location information,light-field images can also record light direction information.Light-field imaging has become an important means of information acquisition due to its unique advantages.It has been widely used in many fields such as stereo matching,image segmentation,target recognition and 3D reconstruction.However,these applications are based on obtaining accurate depth estimation.Therefore,it is particularly important to obtain the depth estimation from the light-field images.In recent years,many excellent depth estimation algorithms have been proposed.But these algorithms have some limitations,such as inaccurate detection of occlusion regions,blurred edges,depth discontinuities of complex texture,and so on.In view of these shortcomings,this paper starts with the optical flow and proposes an improved variable optical flow algorithm.By using this algorithm,we can get more accurate depth estimation results.This paper is mainly based on the variational method of light-field depth flow estimation,our main work and innovations are as follows.Introduced the development and backgrounds of light-field cameras,and systematically analyzed the relationship between optical flow and light-field images and the calculation method.Finally,the status of the optical flow,applications and future trends are discussed.Aiming at the limitation of low accuracy when calculating the depth of light-field based on the traditional optical flow,a depth estimation algorithm based on Lytro camera is proposed.The algorithm mainly consists of three steps:Firstly,the sub-aperture images sequence is obtained through the original light-field images processing,and the optical flow between the center sub-aperture image and the surrounding sub-aperture image is calculated respectively to obtain the global optical flow field.Secondly,detecting the occlusion region by the consistency between the forward and backward flows of the sub-aperture images and removing the occlusion by using the geometric occlusion model,so as to obtain a higher-accuracy initial depth estimation.Thirdly,the voting mechanism is used to fuse the parallax,the histogram is used to calculate the parallax,and the parallax with the largest frequency is selected as the final parallax estimation,using bilinear interpolation with edge preserving property optimizes depth map and thus gets the high precision light-field depth map.Through the computer simulation,the proposed algorithm is compared with the current mainstream methods.Simulation results show that the proposed method has the advantages of low time-consuming and high-precision.