Research On Light Field Image Redirection And Image Enhancement Algorithm

Light field view synthesis is a scientific problem closely related to the redirection of light field images.The existed view synthesis methods are mostly aimed at light field interpolation,and the generated viewpoint range is limited,and the baseline of the light field image cannot be modified.At the same time,when the viewpoint is extrapolated,there is still no reasonable solution to the hole problem generated.Method."Photograph first,focus later" is a major feature of light field images,and light field image refocusing is a hot spot in light field image enhancement algorithms.However,the traditional refocusing algorithm has aliasing phenomenon,which seriously damages the visual effect of the refocusing image.Based on this,this paper proposes a light field image baseline editing algorithm and a light field image new viewpoint generation algorithm,and a light field image refocusing algorithm is proposed to solve the aliasing problem.The main contents include:1)The light field image taken by the handheld light field camera has a short baseline.In order to improve the baseline,a light field image baseline editing algorithm is proposed.Firstly,the light field image is calibrated to obtain the camera parameters,and the disparity map of the light field sub-viewpoint image is calculated.Then use the DIBR algorithm to perform projection transformation on the light field image to complete the preliminary redirection processing.Finally,the proposed deep neural network is used to optimize the target light field image,which mainly includes hole filling and texture repair.Experimental results show that the proposed algorithm can complete the light field image redirection task of baseline editing,and obtain high-quality target light field images.2)Light field refocusing algorithm.The traditional refocusing algorithm based on superposition of sub-viewpoints has serious aliasing phenomenon,and the refocusing method based on light field image reconstruction is too computationally expensive and difficult to improve performance.To this end,a light field image refocusing method based on conditional generation countermeasure network is proposed.The proposed algorithm takes the light field image as input,first calculates the disparity map,and then calculates the required circle of confusion(COC)image from the disparity map.The proposed condition generation confrontation network performs bokeh rendering on the central sub-viewpoint image of the light field according to the COC image,and finally generates a refocused image with the focus plane and depth of field corresponding to the COC image.Compared with the previous algorithm,the method proposed in this paper solves the aliasing Problem,optimize the bokeh effect,and significantly reduce the calculation cost.3)The main body of the light field View Synthesis algorithm is an end-to-end conditional confrontation generation network.The network generator is divided into four parts,the spatial angle feature extraction module,the pose information input module,the fusion feature bottleneck module,and the light field reconstruction module.Among them,we use spatial angle convolution to process the light field macro-pixel image,and integrate the pose information of the target viewpoint into the angle feature.The angle feature is used to calculate the depth,and the spatial feature is used to reconstruct the entire light field.