Research On Text Image Restoration Method

Picture transmission is becoming the most intuitive way that people receive and acquire information in this information age, its clarity deserves the key technology factor of people’s pursuit. But in real life, the text image that people accessed is not clear in many cases. How to recover their important information from blur text image gradually becomes a hot area of research, So it is imperative to find more optimal restoration method of text image.Firstly, this paper introduces the image restoration algorithm theory and frame. Secondly, the exploration and research of non-blind and blindness method of text image restoration were carried out.In the research of non-blind restoration, this paper take text image segmentation smooth and have more step edges characteristics into consideration, the algorithm used should have good edge retention capacity and better denoising effect. Consider the fuzzy text image caused by the movement and noise as a whole, focusing on the sparse reconstruction algorithm part of Fourier transform at first. Secondly, the use of total variation (TV) method to recover blurred text image,through tuning the parameters in this method so as to acquire the better recovery results.On the basis of above,this paper recover the text image use traditional methods and above two methods respectively, the results show that the sparse reconstruction algorithm part of Fourier transform method have good performance for blurred image caused by noise,the total variation have better restoration of blur text caused by the motion and noise, and it is better than the traditional rehabilitation methods.In the research of blind restoration, this paper describes the method that removing unknown blur from the text image. Firstly, it researches the adaptive histogram equalization strong edge retention characteristics based on the characteristics that text image have strong directional and text edges contain important information. Secondly, using Bayesian combined with adaptive histogram equalization methods to find the blur kernel that the image distribution may be implied based on the work of Miskin and MacKay. Given this core, using a standard image deconvolution algorithm for text image reconstruction. The results show that the method is better than other methods after combining. Finally, it summarizes and compares the advantages and disadvantages of algorithms, and the work direction and focus for the future was prospected.