Researches On Text Image Super-resolution Reconstruction Method

High-resolution image can not only give people a better visual experience, but also can provide richer details. Therefore, in the current information age, high-resolution image has a extremely wide range of applications and needs.The collected images may have some flaws, such as edge blur and low recognition in case of the poor situation, low accurate imaging system and human factors. Although the way of enhancing the equipment characteristic can alleviate the problems to some extent, it would not only greatly increase the economic cost, but also be unable to thoroughly eliminate the influence of others factors.The super resolution reconstruction technology has been proved to be a feasible solution. Through the way of the image post process, the technology can make abundant graphic detail and make it easier to distinguish and understand on the basis of the existing conditions, and make the image have a good visual effect. Compared with natural images, the text image has its own unique characteristics, but the general super resolution reconstruction algorithm does not take into account the text feature of the text image, so the processing effect is not ideal for the text image. In this context, some researches are carried out on the reconstruction of text image, the main work includes three parts as follows:1. Analysis on the theory of reconstruction of text image. Firstly, the theoretical basis, the algorithm classification and the research status are introduced. Then the degradation process and observation model are established. Then the basic steps of the super resolution reconstruction algorithm are described. Lastly, several common classical algorithms are analyzed and compared.2. Research on feature extraction technology in the image registration. Firstly, the basic concept and category of the image registration algorithm are discussed. Secondly, Harris, SIFT, SURF are introduced and compared. Then the traditional SIFT algorithm is deeply analyzed, it is found that the original algorithm has a low matching speed, low efficiency and high error rate. To solve this problem, this paper proposes a feature extraction algorithm based on reduced dimension SIFT operator. Finally, the performance of the proposed algorithm is verified by experiments. Experimental results show that the proposed algorithm can effectively reduce the number of mismatch points and matching time and increase the right ratio of matching.3. Research on text image super-resolution reconstruction technique. The paper makes a deep analysis on the projection algorithm, it is found that the unchanged original algorithm threshold will affect the final reconstruction effect. Then combined with the feature that text image grey value shows bimodal distribution, and in the traditional POCS method, the threshold value is not changed, POCS based on adaptive threshold and double-peak envelope algorithm is proposed and implemented. Finally, the performance of the proposed algorithm is verified by experiments. Experimental results show that the proposed algorithm can eliminate noise better, enhance the resolution and improve the image definition.