Research On Structured Measurement And Jointt Reconstruction Algorithm In Multiview Image

With the development of electronic information technology,more and more new applications appear in people's field of vision,such as monitoring system,robot,medical imaging,and satellite imaging.Such applications produce a large number of recorded images of the same scene but different perspectives,namely the multi-view images.The amount of multiview images data is too large,that can often cause network transmission and storage problems.Just relying on updating hardware performance to solve the problems will face huge economic investment.Compressed sensing technology can significantly reduce the sampling rate of the signal,so many scholars put forward solving the problem of multi-view images by this advantage.However,there are still two problems: a large number of the redundancies among the mesurements lead to huge data volume;the quality of image reconstruction needs to be improved.Especially at low sampling rates,details of the reconstruction results are not accurate enough.To solve the above-mentioned problem,this paper proposes two approaches: we proposed a kind of optimal observation method to solve the problem of the redundancies among the mesurements by using correlation between images;considering the correlation between inter-image and intra-image,we proposed a joint reconstruction model to solve the problem of low reconstruction quality.Specific model is as follows:First,for the problem of large amount of observations data,we proposed a structured sampling model for multi-view images based on the inter-image correlation.First,we obtain a structured sampling model according to the correlation between multi-view images,and optimize the measurement matrix;Secondly,we get the measurements by this matrix from the multi-view images.Finally,we preliminarily reconstruct images by using the above-mentioned structured observations.Experiments show that the method greatly reduces the sampling rate by our proposed structured sampling algorithm and the multi-view imagess can get better reconstruction results at the same time.Compared with the traditional sampling,this method is better performance.Second,aiming at the problem of multi-view image reconstruction quality,this paper proposes a joint optimization model to make full use of intra-image and interimage correlation.First,the joint reconstruction method is composed of two parts: adaptive disparity compensated residual total variation and multi-image nonlocal lowrank tensor.The first part proposes the adaptive dynamic image set to reconstruct images,rather than direct use of adjacent images;the second part puts forward the combination of tensor model and the nonlocal low-rank model to further constraint reconstruction.Then,we propose a joint optimization model and an efficient algorithm.Experiments show that,compared with state-of-the-art methods,the proposed scheme obtains dramatically improved visual quality in terms of both subjective and objective quality,especially at low sampling rates.