Research Of Scattering Imaging Based On Optical Information Entropy

Scattering imaging is an image restoration technology that uses speckle pattern with the object information that is disordered by the scattering medium,which has important potential in imaging through opaque layers such as biological tissue,clouds and haze,and turbid water.This dissertation proposes a new method to research scattering imaging that applies the information theory methods of information description and information processing.Firstly,some concepts,such as mutual information and information entropy,were introduced based on the probabilistic interpretation of optical images in scattering imaging,and an information theory model of scattering imaging was established,which contains the source model,channel model and sink model associated to object,scattering medium and camera image,the information transmission model and the maximum entropy image restoration model.Secondly,two different algorithms were given in this dissertation for solving the maximum entropy image restoration model respectively based on the sequential quadratic programming and the alternating direction method.A numerical experiment with 16×16 objects shows that the maximum entropy image restoration method can correctly and effectively recover the object images.The PSNR of retrieved images by sequential quadratic programming is over 17 d B,which means maximum entropy algorithm based on sequential quadratic programming has good anti-noise performance.Inversely,the maximum entropy algorithm based on alternating direction method has a higher computing speed than the former,of which the run time is less than 1 s.Then,a passive scattering imaging experiment system was designed,where a digital micro-mirror device was used as the object,and two distance-adjustable ground glasses were used as the scattering medium.The memory effect range decreases with the increase of distance between two ground glasses.The experimental results show that the maximum entropy method is not limited by the memory effect and can effectively recover object images from memory effect range to non-memory effect range.When the size of sampling points is equal to 7.4 ?m,with the mutual information increasing from 0.1 Nat to 1.8 Nat,as the number of sampling points increases,the PSNR of reconstructed images increases from about 9 d B to about 18 d B.Therefore,the more mutual information a channel transmits,the better image reconstruction quality we obtain.Finally,this dissertation studied the problem of reducing sampling points to achieve sampling optimization in image restoration.From the information transmission model,the influence of the object information entropy on the mutual information between object plane and image plane was analyzed theoretically,and the conclusion is that the mutual information will tend to be a constant as the object information entropy increases.An experiment was designed to verify this conclusion.For a scattering channel with 150 sample points,the mutual information tends to be constant 0.9 Nat as the object information entropy increases.From the experimental results,it is deduced that the minimum number of sampling to recover object image increases with the object information entropy.Furthermore,using the optimal sampling theory based on object sparsity,we have made an consistent conclusions.