Research&Implementation Of Image Super-resolution Reconstruction

Image super-resolution reconstruction is a very important technology in the digital image processing, which reconstruct a image that have higher resolution and richer information details from one or multiple low-resolution images. From it’s definition we can see that the technology include single image super-resolution reconstruction method and multiple images super-resolution reconstruction method. Multiple images super-resolution reconstruction using the existence of sub-pixel displacement and redundant information between multiple low-resolution image sequence to get the reconstructed image. Recently, a lot of scholars focus on the technology. Within scholars’long-term research, the technology has been a very good application in the field of medicine, remote sensing, security monitoring and so on.This paper focuses on the research of single image super-resolution reconstruction method and multiple images super-resolution reconstruction. For single image super-resolution reconstruction method, analyze the simulation results for the algorithm. For the multiple images super-resolution reconstruction, propose a improved algorithm and according to the simulation results to proved the effectiveness of the improvement. The specific work includes the following aspects: Firstly, within the theory of the inversion analysis, this paper use Matlab to simulate this algorithm based on the noise. By comparing the different stages of the reconstructed image de-noising results to demonstrate the necessity to de-noising in the image before reconstruction. Then, with comparative experiments show that there are limitations of existing subjective evaluation and objective evaluation of the reconstructed image, which not fully reflect the reconstruction effect. After researched the noise-based regularized super-resolution image reconstruction method, different picture must use different regularization parameter, so this paper proposed a proper method to get the regularization parameter. Furthermore, based on research of structure-adaptive normalized convolution, through some simulation experiments we can find that the original algorithm has some limitations. So, this paper proposed a improved algorithm, using the adaptive anisotropy. The improved algorithm can effectively improve the quality of the reconstructed image and improve the clarity of the reconstructed image, thereby enhancing the versatility of the original algorithm. Lastly, completed the License plate recognition system based on Matlab. Not only compared the reconstruction results and calculated the objective performance indicators, but also used the proposed algorithm into the license plate recognition technology to improved the recognition rate to verify the effectiveness of the improved algorithm.