Research On Dense Light Field Reconstruction Algorithm Based On Sparse Sampling

Light field imaging technology has become an important tool to obtain the information of the targets in space owing to its multi-dimensional visual perception and expression ability.It can accomplish the tasks of computational refocusing,de-occlusion reconstruction,all-focus synthesis and depth distribution calculation in scene by processing the captured information of light field in space.However,the current light field technology has the problem that it cannot guarantee to obtain high spatial resolution and high angular resolution data at the same time,which restricts the development and application of the technology.At present,researchers in this fields has developed many methods such as increasing the number of detectors,interpolating the rays in angle,synthesizing viewpoint based on geometric depth and frequency replication to solve this problem.Although these methods improve the resolution of the light field to a certain extent,there are still some problems such as low signal-to-noise of the reconstructed images of the light field,high computational complexity of the algorithm,and limited adaptive scope for the scenes.In order to solve the problem that the light field imaging technology cannot capture the high spatial-angular resolution data at the same time,a dense light field reconstruction method based on sparse sampling is proposed.And based on the sparse representation theory,a dense light field reconstruction algorithm framework is established.On this basis,researches on acquisition and characterization of light field,sparsing coding for multi-dimensional data were carried out.The mathematical model for light field dictionary learning and sparse coding were established.A small-scale global light field dictionary was trained to map light field multi-dimensional data to the one in low-dimensional sparse representation domain,the problems of global dictionary training and sparse coding for light field are solved.By further studying the spatial-angular correlation and redundancy,global-local constraint relationship in the multi-dimensional data of the light field,a mathematical model of sparse coding restoration of the virtual angular image of the light field is established,which converts the solution of the virtual angular image into sparse representation of the image using the trained light field dictionary.Then the virtual view images can be reconstruction through image fusion technology.Finally,the target of reconstructing virtual view images of light field is achieved.At last,the experiment is designed to apply the algorithm to the dense light field reconstruction of multiple scenes with sparse sampling.The experimental results show that the proposed method can effectively distinguish between high and low frequency in the scene,recover the complex shadow,illumination changes and the occluded information in satisfying quality.So a conclusion can be arrived that the method in this paper can accomplish the task of reconstruction of dense light field based on sparse sampling.