An Endoscopic Image Sensing And Reconstruction Method Based On The Improved Multi-dictionary Compressed Sensing Framework

In recent years,people’s eating habits have become increasingly bad,and the incidence of gastrointestinal diseases has also increased.Electronic endoscopes are becoming more and more familiar as the most important gastrointestinal diagnostic tools for modern gastroenterology.However,due to various technical problems,the existing electronic endoscope imaging method is not well,which brings great problems to the doctor’s diagnosis.Therefore,it is of great research value and practical significance to study the upgrading and transformation of electronic endoscopes,improve imaging speed and reconstruct imaging accuracy.In this paper,aiming at the similarity between frames and frames in the video stream of electronic endoscopes,the current Compressed Sensing(CS)theory in the field of signal processing is used as the starting point,pointing out the shortcomings in the existing CS image processing framework.Aiming at the shortcomings,an endoscopic image perception reconstruction method based on multi-dictionary improved compressed sensing framework is designed.Under the traditional compressed sensing theory,single dictionary learning requires a one-time use of the entire data set.Dictionary training is time-consuming and robust,so this paper proposes to use a multi-dictionary structure to solve this problem.In addition,the size of endoscopic images tends to be large,including many texture details.The traditional sensing matrix design method can not meet such complex performance requirements.This paper proposes a robust sensing matrix design method based on multi-dictionary average gradient descent.The endoscope reconstruction performance is improved,and finally the image signal is reconstructed and reconstructed based on the multi-dictionary average structure.The specific content of the study is as follows:(1)For the endoscopic image training set,the characteristics of the frame and frame in the real-time video stream are similar.First,in the dictionary learning stage,the larger training set is divided into smaller ones.The sub-training set trains the same size dictionary for the gastrointestinal parts with different detail features.The parallel operation can be used to accelerate the dictionary learning exponentially,and more dictionaries contain more characteristic atoms,and finally reconstruct The stage uses the average weight for weighted reconstruction,which reduces the contingency of signal reconstruction error and improves the reconstruction performance.(2)For the problem that the endoscopic image size is large and contains many texture details,a robust sensing matrix design method with better adaptability to complex multi-variable signals is introduced.This method considers its design in the sensing matrix.The problem of self-energy and the introduction of related regular terms,this paper points out the problem of non-convexity of the cost function of the optimization algorithm after analyzing the optimization function,and innovatively uses the multi-dictionary average gradient descent method to optimize the non-convex cost function.The convergence and reliability of the method are proved by mathematical theory,and the simulation proves that the method can help jump out of the local optimal trap.(3)Using MATLAB simulation software,based on the artificial input signal and the real endoscopic image input signal,the framework proposed in this paper and other excellent image Compressed Sensing processing algorithms are compared horizontally,from norm energy and mean square error.The multi-angles of sensing matrix mutuality,Gram matrix structure and peak signal-to-noise ratio verify the superiority of the proposed algorithm.Finally,the application of other algorithms in other scenarios is carried out.