Research On Super-resolution Reconstruction Algorithms Based On Image Sequence

Within the application domain of image and video, a large number of images and videos are low in resolution in that environment disturbance and limitation of imaging systems introduce drawbacks to resolution. What is more, image and video compression, which is convenient for storing and transmitting image and video, also does harm to resolution of images and videos. Because of lower resolution, people obtain less information from images and videos. Technically, it is very difficult to improve resolution by means of bettering system performance. Moreover, it is not worth doing so because of the imbalance between pay and benefits. Therefore, under the precondition of system hardware not being modified, the research on increasing the resolution of images and videos has become an important branch of image processing.Super-resolution reconstruction technology is an efficient way to increase the resolution of images and videos, which reconstructs one or more high resolution images through signal processing making use of complementary information among multi-frames low resolution images. Recently, this technique has been found widely used in video monitoring, remote sensing, detecting and so on.This paper introduces the primary procedures of super-resolution reconstruction technology systematically, and mainly focuses on following aspects:(1) Motion estimation of images. Super-resolution reconstruction relies on accurate motion estimation between multi-frames, so this paper mainly researches Taylor-series method,optical flow method and block motion estimation, and analyses the performance of algorithms by experiments. The paper adopts a new three steps search block motion estimation algorithm to ensure the precision and improves the steady performance of motion estimation, aiming to meet the requirement of motion estimation's precision and stability of super-resolution reconstruction.(2) Probability and statistics model super-resolution reconstruction algorithm based on MAP estimation. Firstly the objective function is derived and obtained according to the principle of MAP criterion, then the image is deemed as a random field, subjected to certain probability distribution. Finally high resolution is reconstructed after probability and statistics model of image substituted into the objective function. This paper adopts Markov priori model and analyses the defects of this model in reserving the edges of images, and then an improved method is put forwarded based on the work of predecessors. Experiments prove that the performance of edge preservation of images is improved effectively using the improved method.(3) Variance regularization super-resolution reconstruction algorithm based on Lp norm. Firstly a frame for super-resolution reconstruction is constructed, which includes data distortion term and regularization term. Secondly, this article introduces a super-resolution reconstruction algorithm based on bilateral total variance model. Then an algorithm of edge-enhanced nonlocal model super-resolution reconstruction based on nonlocal algorithm is proposed, analyzed and compared with improved Markov model and bilateral total variance model. Finally, objective function for reconstruction based on L1 form is applied after comparing L1 form with L2 form. Experiments prove that the algorithm this paper proposed performs well not only in restraining the noise, but also in preserving the edges information of images.