Research On Super-Resolution Display Algorithm Of Text Images

Due to the limitation of pixel size and imaging equipment costs,the acquired text images usually have the feature of low resolution.The images often appear blurred and unclear when performing high-resolution display,which affects human visual appreciation and text understanding.As an effective software solution,text image super-resolution is to reconstruct a high-resolution text image from one or more low-resolution text images.However,the existing methods tend to cause text blurring,incomplete edges and missing details.To solve this problem,according to text image priors,this thesis is successively devoted to the research on the super-resolution reconstruction of printed and handwritten text images.For printed text images,this thesis proposes an edge-driven super-resolution algorithm.First step of this algorithm is detecting the edge of the input low-resolution image.Secondly,based on the edge image,the pixels of the input low-resolution image are divided into two categories: edge pixels and non-edge pixels.Then,bicubic interpolation and autoregression model are used for high-resolution image reconstruction for edge pixels and non-edge pixels respectively,while the structural self-similarity between the image blocks is used for improving the autoregressive model.Finally,the reconstructed super-resolution text images are acquired by collecting edge pixels and non-edge pixels.The experimental results show that the proposed algorithm can ensure the continuity and sharpness of the text edges,effectively suppress the distortion phenomena such as edge blurring and sawtooth distortion of the reconstructed images,and obtain better super-resolution results than the existing classical methods.In this thesis,a super-resolution algorithm based on edge residual compensation is proposed for handwritten text images.Initially,the algorithm uses the above edge-driven super-resolution algorithm to estimate the target high-resolution image.Secondly,the current estimation of the target high-resolution image is down-sampled and combined with bilateral filtering to obtain the same size of the input low-resolution image,while the structural contrast between the image blocks is used as a prior to improve the joint bilateral filtering algorithm.Then,the edge of the sampled image and the input low-resolution image are detected,and the residual error is calculated based on the acquired edge images.Nextly,the above edge-driven autoregressive model is used to interpolate the edge residual,which is compensated to the current estimate of the target high-resolution image by using backprojection.Finally,to reconstruct a super-resolution text image,backprojection is used for edge compensation with multiple iterations until convergence.The experimental results show that the proposed algorithm can effectively suppress the text edge blurring and reconstruct more text details.