Research On Light Field Image Denoising An D Spatial Super-resolution

As a new imaging technology,light field imaging breaks through the limitations of traditional two-dimensional imaging,it can obtain the direction information of the light while recording the position information of the light.This additional information can make the light field widely used,such as depth estimation,object detection,light field 3D rendering,3D reconstruction,etc.However,due to the limitation of the hardware conditions of the light field camera and the interference of external factors such as the environment,light,and jitter during the shooting process,the obtained light field images often have low resolution and different levels of noise pollution,which is difficult to meet the needs of practical applications.For this problem,the methods of denoising and super-resolution of light field are studied in this paper.The main work is as follows:First,Aiming at the problem of noise in light field image acquisition,we study the LFBM5 D light field denoising algorithm and improves the LFBM5 D algorithm.In the original algorithm,fixed image block size is used for matching filtering,which does not consider the problem that the optimal image block size of different shape component regions is different in the block matching filtering algorithm.Combined with image block classification,the image blocks of smoothing,texture and edge are matched and filtered with different image block sizes.The basic estimation and final estimation of the original algorithm are divided into four sub-stages: smooth block denoising,texture block denoising,edge block denoising,and overall denoising.In addition,in order to improve the block matching accuracy,SSIM is introduced when Calculate similarity.Experimental results show that the denoising effect of the improved method is better than the original method in both subjective and objective indicators.Second,Aiming at the problem of low spatial resolution in light field image acquisition,we study the light field super-resolution method and improve the single image super-resolution algorithm based on wavelet domain dictionary learning,then combine it with the light field super-resolution to realize the light field super-resolution reconstruction.In the original algorithm,the assumption that the sparse representation coefficients of high-resolution and low-resolution images under two redundant dictionaries are the same is too ideal,which leads to the weak learning problem between the sparse representation coefficients.The semi coupled dictionary learning mechanism is introduced to establish the sparse representation mapping relationship of highresolution and low-resolution images by learning a set of mapping matrices,so as to enhance the reconstruction quality.Experimental results show that the improved method performs better in subjective vision and objective indicators than the original method.The improved algorithm is further combined with the light field superresolution framework to realize the super-resolution reconstruction of light field images.Compared with the current mainstream light field super-resolution algorithms,the experimental results show that the proposed method has certain competitiveness.