Research On Blurred Text Image Restoration

With the increasingly wide range of digital image processing application, text image processing has become an important research area, and fuzzy text image restoration is of great practical value and social significance. During the practical application, the blurred text images are often with little of prior knowledge, so nothing but blind image restoration algorithm can restore them. As a kind of special occasion ones, text images have their unique nature. According to the characteristics of the text images, the current advanced blind image restoration algorithms can be improved in order to obtain better recovery results.In this paper, the basic model of blind restoration algorithms are researched and the common ones are classified and summarized. Their general idea is as follows:At first, the corresponding mathematical model is established based on the process of image degradation; At second, a reasonable estimate of the original image is reached by solving out the inverse problem. In addition, the characteristics of the text images get analysed deeply, that is bimodal characteristic about histogram, setting the stage for the presentation of the new blind image restoration algorithm based on text image. Focuses on the classic blind deconvolution algorithms and the advanced NAS-RIF algorithm, and improve them respectively. The improved blind deconvolution algorithm uses the new spatial constraints, and the improved NAS-RIF algorithm modify the cost function, which merges the bimodal characteristic of text images as a limiting condition. Through the comparative analysis of abundant experiments, it is clear that, when processing text images with the same blur length, from the results of image restoration perspective, the improved algorithm has obvious advantages than the traditional one in the same class, while without ringing phenomenon the improved NAS-RIF algorithm can get higher text recognition than the improved blind deconvolution algorithm. However, they both are sensitive to noise and need to be further advance. Finally, the advantages and disadvantages of both algorithms are summarized and analyzed and the direction of future work is shown.