Image Processing And Application Of Light Field Based On Deep Learning

Light field imaging technology can record more dimensional information of the light field.Compared with RGB images,it contains richer characteristic information and has broad application prospects and research value.Meanwhile,its multi-dimensional data has also brought challenges to research in many fields.However,deep learning methods have good performance in processing such high-dimensional and large amounts of data.This makes it possible to use deep learning methods to process light field data.The project aims to use deep learning technology to study the processing and application of light field images,to achieve 3D plane detection and target detection,and to provide a reference for post-processing work.First of all,according to the characteristics of light field image multi-aperture view,the thesis designs a 3D plane detection method of light field images based on deep learning.The key is to use multiple sub-aperture images of the light field as the input of deep neural network for feature fusion.The output is the result of plane detection and depth estimation.Furthermore,the thesis improves the Mask R-CNN algorithm and designs a target detection architecture suitable for multi-image input.It takes four images with different polarization directions as network inputs,and then fuses the features to detect the target.The test results show that the average precision of the architecture is 8% higher than that of the traditional method,and the miss rate of the architecture is 18% lower than that of the traditional method.Finally,aiming at the problem that the target detection system is greatly affected by interference in the complex environment,the thesis designs a multi-dimensional light field imaging target detection method.First,the multi-dimensional image is acquired according to the principle of light field imaging.Then the multi-dimensional image is extracted and fused using the designed target detection architecture to detect the target.The final test results show that the multi-dimensional light field imaging target detection method improves the ability of the target detection system to resist complex background interference.The average precision of the method is 14% higher than that of the traditional method.