Research On Two-dimensional Code Restoration Algorithm Based On Sparse Regularization And Deep Learning

With the rapid development of Internet and smart devices,QR code images play an extremely important role in the daily life and business activities.However,the process of image acquisition is often affected by many factors,resulting in the loss of the details and key information of the image.Image restoration proposed to improve the image quality is a typical ill-conditioned inverse problem,whose unknown variables are much more than the known,so it is hard to obtain a stable and reliable solution.Therefore,the current restoration algorithms usually use some hypothesis or prior knowledge in the degraded system as the constraints of the solution model,and minimize the objective function to get the optimal approximate solution and recover the key information.The restoration of QR code images has extremely wide application and research value in various aspects such as commodity payment,information exchange,website registration,commercial advertising,etc.Faced with the fact that the binary image represented by QR code is widely used but the related restoration research is rare,this paper mainly discusses the restoration algorithm of the 2D code image:Firstly,the research background,significance and the research status at home and abroad of image restoration are briefly introduced.At same time,this paper summarizes the existing classical image restoration algorithms,and based on the research status of binary image restoration algorithm,proposes the necessity of exploring a more efficient 2D image restoration algorithmSecondly,this paper introduces the process and model of image degradation/recovery,and summarizes the classical image restoration algorithms,including filtering method and iterative regularization method.It also points out the advantages and disadvantages of different methods,and finally gives the subjective and objective image quality evaluation index.Then,a two-dimensional code image restoration algorithm based on binary constraint and sparse regularization is proposed.By observing and analyzing the grayscale and gradient histogram of the clear/blurred binary image,the binary constraint,the₀L norm of pixels and gradients is added as a regularization constraint to the minimized restoration model.The whole process is divided into two steps:blur kernel estimation and image estimation.By continuously alternating to minimize the optimal solution of the two steps,the estimated values of the clear image and the blur kernel are finally obtained.Finally,a two-dimensional code image restoration algorithm based on convolutional neural network is proposed.By verifying the influence of network depth and different loss functions on the network effect,the appropriate network depth and loss layer are selected.According to the characteristics of the binary image,the binary constraint is added to the loss function.At last,the experimental results show that the proposed algorithm achieves optimal results both in subjective visual perception and in objective numerical index.Compared with the traditional iterative method,the proposed algorithm has less time consumption,is more robust to noise,and has better generalization performance.