The Research Of Super-resolution Of Document Image Based On POCS Algorithm

In many research areas, analysis of the image details is very important. Thehigher resolution the image has, the richer details it contains. Therefore, thereconstruction of super-resolution has already become a hotspot in the researchfield of image visualization.At present, there is a great application demand of the document imagereconstruction in certain cases. Take these as examples: it can make theextracted license plate number more clear which has been magnified during along-distance recognition, so that it turns out a safer traffic monitoring; it canhighlight faded and fuzzy character strokes to make it possible to restore theoriginal appearance when saving old books and relics; it can help reconstructclear inscription to make the research much easier when studying history. Manyexisting reconstruction algorithms are for the natural image. For the documentimage has different texture features to natural ones, the reconstruction effect ofexisting algorithms is not ideal. Therefore, it is necessary to build areconstruction algorithm which has remarkable effect with the text features of document image.The paper has a further study of the super-resolution reconstruction ofdocument image. It focuses on the image registration and reconstructionalgorithm, lists relevant optimization methods for the inadequacy of traditionalalgorithm, and realizes the ancient inscription application. The paper centers ondocument image reconstruction to launch the research from the followingaspects.Firstly, it introduces scientific background and current progress of thesuper-resolution reconstruction of document image at home and abroad,elaborates specific application and theoretical basis of the study, establishesdegradation model of document image, details the super-resolutionreconstruction with three steps, and presents the reconstruction techniques inclassification.Secondly, the paper describes the image registration algorithm, and focuseson the research and improvement of SIFT operator in feature point extraction. Itpreprocessed the document image on the basis of the original algorithm toimprove the image contrast. At the same time, it reduced dimension of theoriginal algorithm to save a lot of memory space of SIFT operator, accelerate thefeature point extraction rate and improve the accuracy of matching. The papercombines the dimension reduction with the existing spatial domain method ofimage reconstruction and applies it to super-resolution reconstruction of videoimage sequence acquired, which improves the reconstruction effect. Finally, it highlights and improves POCS of spatial domain method tomake it more suitable for document image reconstruction. In registration, it usesimproved SIFT registration method. In reconstruction, it corrects estimates withthe priori constraint of unique bimodal characteristics of document image, whichis better-oriented compared with the original algorithm. In threshold selection, itadds uncertainty factor to the degenerate model and transfers fixed threshold oforiginal algorithm to adaptive threshold, which improves the algorithmrobustness and image reconstruction effect. Then, the paper applied theimproved algorithm to the ancient inscription image reconstruction. Text linesare prominent in reconstructed image and the noises caused by the line shadowsare eliminated on the whole. Compared with the traditional algorithm, thereconstruction effect in the paper is more ideal, which proves that the improvedalgorithm has higher feasibility in practical applications.